The 30th Special CU – af Seminar 2022

The 30th Special CU-af Seminar 2022 September 2, 2022 4.2 Å, which represents the second solvation shell around the Zn2+. This result is similar to other concentrations, as shown in Figure 6 (b)-(d). When the system concentration increased, the Zn2+-Oanion coordination number increased as expected. This is due to increasing in the number of zinc salt molecules in the systems. It is apparent that adding 0.1 M MnSO4 resulted in lower coordination between Zn2+ ions and anions, which indicates weak interaction of salt. From the results in Figures 5 and 6, which are consistent with the structural snapshots obtained from the MD simulations as shown in Figure 7. It shows the structural snapshots of Zn2+ ions of ZnSO4 with and without 0.1 M MnSO4 at 0.5 M and 2.0 M. Figure 7: The structural snapshots of Zn2+ ions of (a) 0.5 M ZnSO4, (b) 0.5 M ZnSO4 + 0.1 M MnSO4, (c) 2.0 M ZnSO4, and (d) 2.0 M ZnSO4 + 0.1 M MnSO4. Transport properties Figure 8: (a) Mean squared displacement of Zn2+ ions and SO42- anions in the electrolyte of 0.5 M and (b) diffusion coefficients obtained from the simulated systems of ZnSO4 and ZnSO4 + 0.1 M MnSO4. Figure 8 shows mean squared displacement and self-diffusion coefficient obtained from simulations of ZnSO4 and ZnSO4 + 0.1 M MnSO4. In Figure. 8 (a), the plots become linear beyond 1 ns, indicating that the Zn2+ get into diffusive regime as early as 1 ns for given simulation conditions. The self-diffusion coefficients of aqueous ZnSO4 as a function of concentration were simulated from mean square displacement of Zn2+ ions and SO42- anions and plotted in Figure. 8 (b). For both systems, ZnSO4 and ZnSO4 + 0.1 M MnSO4, with an increase in salt concentration, the diffusivity of Zn2+ ions and SO42- ions decreases. It appears that systems 36

The 30th Special CU-af Seminar 2022 September 2, 2022 of high concentrations in which strong coordination between Zn2+ ions and anions are obtained, resulted in slower dynamic of both ions. Another important reason for the decreasing in these properties could be the formation of ion pairs and complexes that are essentially less mobile. This behavior has already been noticed in lithium ion systems[21,28]. Adding MnSO4 resulted not only in stronger coordination between Zn2+ ions and water, but also in weaker coordination between Zn2+ ions and anions. However, the trends observed in the diffusion coefficient for ZnSO4 solutions with MnSO4 lower than those without MnSO4. This is consistent with more congested systems having less available free space. Ionic conductivity Figure 9 shows impedance of ZnSO4 with and without 0.1 M MnSO4 in the range of 1000 kHz to 0.5 mHz. The impedance data can be used to evaluate the ionic conductivity. Figure 9: Impedance of (a) ZnSO4 and (b) ZnSO4 with 0.1 M MnSO4 for a range of ZnSO4 concentration. Figure 10 shows ionic conductivity as a function of concentration obtained from MD simulations and experiments. With an increase of the concentration of ZnSO4 electrolyte, the diffusivity of Zn2+ ions decreased, leading to lower the ionic conductivity. The highest ionic conductivity obtained from the simulation is at 0.5 M ZnSO4 concentration, for ZnSO4 with and without 0.1 M MnSO4 additive. The ionic conductivity initially shows an increasing trend up to a certain concentration, beyond which it decays rapidly, especially at 2.0 M of ZnSO4 + 0.1 M MnSO4. The explanation for this may lie in the possible increase of unstable ion-paired anions and/ or solubility limit with increasing concentration[29]. However, the experimental ionic conductivity reaches its maximum value at 1 M ZnSO4 concentration. Although, the ionic conductivity from MD predictions is much higher than the experimental value. It is perhaps due to experimental are complex system that are resistant of copper wire and purity of aqueous electrolytes. Figure 9: Ionic conductivity as a function of concentration obtained from MD simulations and experiments. 37

The 30th Special CU-af Seminar 2022 September 2, 2022 Conclusion The MD simulation study of the ZnSO4 and ZnSO4 + 0.1 M MnSO4 systems for a range of concentrations: 0.1, 0.5, 1.0, and 2.0 M can capture the effect of concentration on the structural and dynamic properties of the electrolyte solution. The results showed the Zn2+ ions were coordinated by six water molecules at 0.1 M ZnSO4 with and without 0.1 M MnSO4 at the radial distance of 2 Å. The coordination number decreased with an increase in the salt concentration. At high concentration, i.e., 2.0 M, the addition of MnSO4 resulted in better coordination between Zn2+ and water molecules, indicating that Zn2+ ions have a good solvation structure. Association of ions between Zn2+ and Oanion (i.e., SO42-) was also observed at 2 Å, suggesting that Zn2+ ions interact closely to SO42- anions. This is due to strong electrostatic force. The coordination number increased with an increase in the salt concentration. This may affect the stability of electrolyte if ion-paired anion becomes unstable in the solution. The addition of 0.1 M MnSO4 resulted in less coordination between Zn2+ ions and anions, as well as weaker salt interaction. As the concentration increased for both systems, the diffusivity of both Zn2+ ions and SO42- ions and the ionic conductivity of the solution decreased. The ionic conductivity increased with concentration to reach its peak value of about 0.5 M concentration and then decreased. As compared with experimental results, the ionic conductivity is maximum at 1 M ZnSO4 concentration. At low concentrations, solvent separated ion pairs are dominant. The solution contains more contact ion pairs and aggregates as the concentration increases. According to the analysis of various radial distribution functions and coordination number, the number of free SO42- ions in the solution is depleted with an increase in salt concentration. As a result, salt solvent complexes such as contact ion pairs and aggregates can develop. Lower mobility of the aggregates due to their size and smaller amount of charge carried per ion is the reason for the decrease in the ionic conductivity. A way to reap the benefits of concentrated electrolytes is to identify a combination of solvent, salt, and cosolvent that aids in achieving better ion mobility and a high count of solvent-separated ion pairs at higher concentration. The function of regularly used cosolvents in the performance of concentrated electrolytes is being investigated. References 1. Wang, Z.; Tan, R.; Wang, H.; Yang, L.; Hu, J.; Chen, H.; Pan, F., A Metal–Organic- Framework-Based Electrolyte with Nanowetted Interfaces for High-Energy-Density Solid-State Lithium Battery. Advanced Materials 2018, 30, 1704436. 2. Guo, X.; Ding, Y.; Xue, L.; Zhang, L.; Zhang, C.; Goodenough, J. B.; Yu, G., A Self-Healing Room-Temperature Liquid-Metal Anode for Alkali-Ion Batteries. Advanced Functional Materials 2018, 28, 1804649. 3. Jia, H.; Chen, C.; Oladele, O.; Tang, Y.; Li, G.; Zhang, X.; Yan, F., Cobalt Doping of Tin Disulfide/Reduced Graphene Oxide Nanocomposites for Enhanced Pseudocapacitive Sodium-Ion Storage. Communications Chemistry 2018, 1. 4. Jia, H., et al., Electrospun Kraft Lignin/Cellulose Acetate-Derived Nanocarbon Network as an Anode for High-Performance Sodium-Ion Batteries. ACS Applied Materials & Interfaces 2018, 10, 44368-44375. 5. Li, H.; Yang, H.; Sun, Z.; Shi, Y.; Cheng, H.-M.; Li, F., A Highly Reversible Co3s4 Microsphere Cathode Material for Aluminum-Ion Batteries. Nano Energy 2019, 56, 100-108. 38

The 30th Special CU-af Seminar 2022 September 2, 2022 6. Su, D.; McDonagh, A.; Qiao, S.-Z.; Wang, G., High-Capacity Aqueous Potassium-Ion Batteries for Large-Scale Energy Storage. Advanced Materials 2017, 29, 1604007. 7. Liu, S.; Chen, X.; Zhang, Q.; Zhou, J.; Cai, Z.; Pan, A., Fabrication of an Inexpensive Hydrophilic Bridge on a Carbon Substrate and Loading Vanadium Sulfides for Flexible Aqueous Zinc-Ion Batteries. ACS Applied Materials & Interfaces 2019, 11, 36676-36684. 8. Cheng, Y., et al., Highly Reversible Zinc-Ion Intercalation into Chevrel Phase Mo6s8 Nanocubes and Applications for Advanced Zinc-Ion Batteries. ACS Applied Materials & Interfaces 2016, 8, 13673-13677. 9. Zhang, N.; Cheng, F.; Liu, J.; Wang, L.; Long, X.; Liu, X.; Li, F.; Chen, J., Rechargeable Aqueous Zinc-Manganese Dioxide Batteries with High Energy and Power Densities. Nature communications 2017, 8, 1-9. 10. Pan, H.; Shao, Y.; Yan, P.; Cheng, Y.; Han, K. S.; Nie, Z.; Wang, C.; Yang, J.; Li, X.; Bhattacharya, P., Reversible Aqueous Zinc/Manganese Oxide Energy Storage from Conversion Reactions. Nature Energy 2016, 1, 1-7. 11. Palaniyandy, N.; Kebede, M. A.; Raju, K.; Ozoemena, K. I.; le Roux, L.; Mathe, M. K.; Jayaprakasam, R., Α-Mno 2 Nanorod/Onion-Like Carbon Composite Cathode Material for Aqueous Zinc-Ion Battery. Materials Chemistry and Physics 2019, 230, 258-266. 12. Chamoun, M.; Brant, W. R.; Tai, C. W.; Karlsson, G.; Noréus, D., Rechargeability of Aqueous Sulfate Zn/Mno2 Batteries Enhanced by Accessible Mn2+ Ions. Energy Storage Materials 2018, 15, 351-360. 13. Cannon, W. R.; Pettitt, B. M.; McCammon, J. A., Sulfate Anion in Water: Model Structural, Thermodynamic, and Dynamic Properties. The Journal of Physical Chemistry 1994, 98, 6225-6230. 14. Panteva, M. T.; Giambaşu, G. M.; York, D. M., Force Field for Mg2+, Mn2+, Zn2+, and Cd2+ Ions That Have Balanced Interactions with Nucleic Acids. The Journal of Physical Chemistry B 2015, 119, 15460-15470. 15. Berendsen, H. J. C.; Grigera, J. R.; Straatsma, T. P., The Missing Term in Effective Pair Potentials. The Journal of Physical Chemistry 1987, 91, 6269-6271. 16. Parrinello, M.; Rahman, A., Polymorphic Transitions in Single Crystals: A New Molecular Dynamics Method. Journal of Applied Physics 1981, 52, 7182-7190. 17. Parrinello, M.; Rahman, A., Crystal Structure and Pair Potentials: A Molecular-Dynamics Study. Physical Review Letters 1980, 45, 1196-1199. 18. Van Der Spoel, D.; Lindahl, E.; Hess, B.; Groenhof, G.; Mark, A. E.; Berendsen, H. J. C., Gromacs: Fast, Flexible, and Free. Journal of Computational Chemistry 2005, 26, 1701-1718. 19. Darden, T.; York, D.; Pedersen, L., Particle Mesh Ewald: An N·Log(N) Method for Ewald Sums in Large Systems. The Journal of Chemical Physics 1993, 98, 10089-10092. 20. Essmann, U.; Perera, L.; Berkowitz, M. L.; Darden, T.; Lee, H.; Pedersen, L. G., A Smooth Particle Mesh Ewald Method. The Journal of Chemical Physics 1995, 103, 8577-8593. 21. Ravikumar, B.; Mynam, M.; Rai, B., Effect of Salt Concentration on Properties of Lithium Ion Battery Electrolytes: A Molecular Dynamics Study. The Journal of Physical Chemistry C 2018, 122, 8173-8181. 22. Haberler, M. In Computing the Static Conductivity of Ionic Liquids, 2008. 23. France-Lanord, A.; Grossman, J. C., Correlations from Ion Pairing and the Nernst-Einstein Equation. Physical Review Letters 2019, 122, 136001. 39

The 30th Special CU-af Seminar 2022 September 2, 2022 24. Benavente, J., Electrochemical Impedance Spectroscopy as a Tool for Electrical and Structural Characterizations of Membranes in Contact with Electrolyte Solutions. In Recent Advances in Multidisciplinary Applied Physics, Méndez-Vilas, A., Ed. Elsevier Science Ltd: Oxford, 2005; pp 463-471. 25. Mansfeld, F., Electrochemical Impedance Spectroscopy (Eis) as a New Tool for Investigating Methods of Corrosion Protection. Electrochimica Acta 1990, 35, 1533-1544. 26. Xu, M.; Zhu, T.; Zhang, J. Z. H., Molecular Dynamics Simulation of Zinc Ion in Water with an Ab Initio Based Neural Network Potential. The Journal of Physical Chemistry A 2019, 123, 6587-6595. 27. Sun, W., et al., A Rechargeable Zinc-Air Battery Based on Zinc Peroxide Chemistry. Science 2021, 371, 46-51. 28. Mynam, M.; Ravikumar, B.; Rai, B., Molecular Dynamics Study of Propylene Carbonate Based Concentrated Electrolyte Solutions for Lithium Ion Batteries. Journal of Molecular Liquids 2019, 278, 97-104. 29. Rajput, N. N.; Qu, X.; Sa, N.; Burrell, A. K.; Persson, K. A., The Coupling between Stability and Ion Pair Formation in Magnesium Electrolytes from First-Principles Quantum Mechanics and Classical Molecular Dynamics. Journal of the American Chemical Society 2015, 137, 3411-3420. 40



Development of polyacrylonitrile/bio-related polyurethane electrospun fiber mats as separator in Zn-ion battery Manunya OKHAWILAI

The 30th Special CU-af Seminar 2022 September 2, 2022 Development of polyacrylonitrile/bio-related polyurethane electrospun fiber mats as separator in Zn-ion battery Manunya OKHAWILAI1* Abstract This research aims to fabricate polyacrylonitrile (PAN)/bio-based polyurethane (bio-based PU) separator for Zn-ion battery. PAN/bio-based PU electrospun fiber mats at 75/25 were prepared via electrospinning technique using Taguchi experimental design with three parameters and four levels. It was found that all fiber mats illustrated smooth and continuous fiber. The properties of the electrospun fiber mats were optimized in accordance to grey relational method. Furthermore, the fiber mats prepared from 14 wt% polymer concentration, 25 kV of applied voltage, and 16 cm of distance from tip to collector were determined to be the most suitable fabrication condition exhibiting an ionic conductivity of 3.11 mS/cm, tensile strength of 44.16 MPa, and electrolyte uptake of 1,971%. Thus, the PAN/bio-based PU electrospun fiber can be considered a promising candidate as a separator for Zn-ion battery. 1Metallurgy and Materials Science Research Institute Chulalongkorn University Bangkok, Thailand 43

The 30th Special CU-af Seminar 2022 September 2, 2022 Introduction and Objectives Over the past few decades, lithium ion batteries (LIBs) have attracted considerable attention as power sources for portable electronic devices and electric vehicles owing to their high energy density, light weight, long cycling lifespan, and eco-friendliness. However, some details of LIBs, high cost and unsafety, are main reason that make other batteries system still be developed in the parallel time. Zinc ion batteries (ZIBs) is one of interesting batteries system because of high theoretical volume capacity of metal Zn anode, nontoxicity, mild electrolyte, low cost and high abundance[1-3]. Usually, batteries consist of two electrodes i.e. cathode and anode, an electrolyte, and a separator. Separator is one of the most important and indispensable part of the battery having the function of 1) separate cathode and anode, 2) support ion transportation between two electrodes, 3) suppress electron movement in the cells, and 4) prevent dendritic formation at the electrode[4]. For LIBs, commercial separators are mainly produced from polyolefin and fiber glass. Their morphologies are illustrated in Figure 1 [5]. To the best of our knowledge, however, there are limit publications or state of arts for separator of ZIB where the mechanism during charge/discharge is different from LIB. Figure 1: SEM images commercial LIB separator[5]. One most important property of separator is its high porosity which is reported that provide high ionic conductivity by supporting movement of metal ion[6]. Wet and dry processes are applied to polymer film to provide high porosity[7]. Electrospinning is an effective technique providing such relatively high porosity and surface area to weight ratio. Electrospinning is a manufacturing technique for production of electrospun fiber mats using electrostatic driven process. This technique can provide the diameter size of ten of nanometres to a few of micrometres. This technique is not only for academic research but also widely use in commercials product application. The popularity of the electrospinning technique has allowed multiple technologies such as food encapsulation, insulating materials, energy conversion and storage, air and water filtration as illustrated in Figure 2. 44

The 30th Special CU-af Seminar 2022 September 2, 2022 Figure 2: Application of fiber mats produced via electrospinning technique[8]. The electrospinning equipment consists of 1) applied voltage to produce charges, 2) syringe pump to control flow rate of polymer solution, 3) syringe filled with polymer solution, and 4) collector to collect fiber mats. The morphology of the obtained fiber mats depends strongly on 1) material parameters i.e. polymer solution concentration, viscosity, and volatility of solvent and 2) processing parameters i.e. applied voltage, distance from tip to collector, and humidity. To study effect of those parameters on relevant properties of the fiber mats, design of experiment (DOE) based on Taguchi method is applied. The optimization of parameters to provide the most of suitable properties of the fiber mats is studied through Grey relational analysis. Thermoplastic polyurethane (TPU) with ether bond is considered as a remarkable flexible material. TPU possesses two-phase microstructure, i.e. soft and hard segments. The soft segment composes of polyhydric alcohols (polyol) with ether bond promoting transportation of metal ions in the electrolyte[9]. The polyol can be classified into polyester polyol and polyether polyol. The hard segment mainly refer to diisocyanate that interact with neighboring ones through π-π interactions as well as hydrogen bondings between the urethane groups[10]. The hard segments also generate physical cross-link points to keep good strength and film forming property[11]. Various types of diisocyanate are used including toluene diisocyanate (TDI), hexamethylene diisocyanate (MDI), and diphynylmethane diisocyanate (MDI). Due to the environmental concern of using petroleum-based polymer, bio-based polyurethane has synthesized usually from bio-based polyol for example caprolactone polyol, and castor oil etc. TPU as separator membrane was reported to provide high ionic conductivity facilitating from the soft segments which do not form ionic clusters after being dissolved in an alkali metal[11]. The copolymers of TPU with other polymers have been developed for example polyethylene oxide (PEO), poly(vinylidene fluoride) (PVDF), and poly(vinylidene fluoride-co-hexafluoropropylene) (PVDF-HFP) etc. to look for a novel material to be compatible and improve TPU properties having a remarkable electrochemical stability. TPU electrospun membrane for LIB was prepared[12]. It was reported that the tensile strength of electrospun TPU was relatively low i.e. 2.4 MPa. After subjected to heat treatment at 170 °C for 60 min in an oven, the membrane showed a large dimensional change which caused a thermal shutdown in the cell. The blending of TPU with high performance polymers 45

The 30th Special CU-af Seminar 2022 September 2, 2022 is alternative technique to improve the pure polymer properties. Polyacrylonitrile (PAN) is one of the polymers which play an important role for using in battery separator application because of its viscoelasticity and excellent resistance to oxidative degradation[13]. PAN membranes also exhibit high thermal stability, excellent electrochemical stability, and good compatibility with electrodes. Moreover, PAN can act as a polymer matrix to help maintain the liquid electrolyte and participate in Li ion transport due to the interaction between the Li ion and the CN group of PAN. The application of PAN separators also minimized Li dendrite formation, which is form during charge/discharge process, due to the viscoelastic of PAN. However, PAN separator often suffer from electrolyte leakage during long-term storage .[14] Another concern points of using PAN membrane is the high shrinkage at a temperature of 150 °C and poor mechanical strength which limit applications where high strength and flexibility are needed[15, 16]. Objectives 1. To synthesis bio-related polyurethane from caprolactone polyol and bio-based diisocyanate 2. To study effect of processing and material parameters on morphologies of polyacrylonitrile/ bio-related polyurethane electrospun fiber mats using Taguchi and Gray Relational Analysis 3. To characterize physical, thermal, mechanical and electrochemical properties of the obtained electrospun fiber mats as separator for Zn-ion battery Methods Bio-related urethane prepolymer using caprolactone diol with Mw of 2000, bio-based diisocyanate, ethylene glycol chain extender at mol ratio of 1:2:1 and dibutyltindilaurate (DBLT) catalyst was synthesized. The synthesis route is shown in Figure 3. Bio-related polyurethane was then mixed with polyacrylonitrice (PAN) using dimethylformamide solvent. The membrane fiber mats were prepared by electrospinning technique (see Figure 4.). The experimental design using Taguchi method coupled with Grey relational analysis with 3 factors of solution concentration, applied voltage and distance from tip to collector by 4 levels was carried out. The factors and levels are shown in Table 1. The morphology of fiber mats was characterized by scanning electron microscope (SEM). The average fiber diameter was from 100 measurements by Image J. The porosity was measured using n-butanol uptake and calculated according to Eq.(1). Figure 3: Synthetic route of bio-based PU 46

The 30th Special CU-af Seminar 2022 September 2, 2022 Figure 4: Electrospinning setup. Table 1: L16 (43) orthogonal array. (1) Where Wd and Ww are the dry and wet weight of the fiber mats after soaking in n-butanol, respectively. is the density of n-butanol, and is the volume of the dry mats. The electrolyte uptake of the fiber mats was determined using Eq.(2). 47

The 30th Special CU-af Seminar 2022 September 2, 2022 (2) where Wd and Ww are the weight of the PAN/bio-based PU electrospun fiber mats before and after immersing the sample in the liquid electrolyte for 1 h, respectively. The wettability of the fiber mats by an electrolyte was observed with electrolyte droplet placed on their surfaces and confirmed by in-house contact angle analyzer. The electrospun fiber mats were dried in a vacuum overnight to remove all moisture. The contact angle measurements were conducted within 5 sec often placing one drop of electrolyte on the samples. The tensile strength were characterized at room temperature using a universal testing machine (Tinius Hitachi, model 5ST) according to ASTM D882. The test was carried out with a gage length of 50 mm and a crosshead speed of 50 mm/min. A potentiostat/galvanostat (PSTrace4 Palm Sens) was utilized for investigating the electrochemical properties. The measurements were performed on an applied 10 mV AC potential from 1 MHz to 1 Hz. The Zn/separator/Zn cell was constructed by inserting the fiber mats between blocking electrodes made of stainless steel. The ionic conductivity could be worked out according to Eq.(3). (3) Taguchi method containing three factors and three levels was used to optimize the conditions of UAE process in terms of single-response and multiple responses. The conditions and their variation levels are shown in Table 2. The signal-to-noise ratio (S/N) was used to evaluate the effect of each parameter level for single-response optimization with the help of ANOVA. The S/N ratios were classified into three classes, namely, 1.nominal-the-better, 2.smaller-the-better, and 3.larger-the-better, which were applied for optimization[17]. In this study, all responses including pore size and porosity were minimized for the “smaller-the- better class,” whereas strength, electrolyte uptake, and ionic conductivity were maximized for the “larger-the-better class.” S/N ratio was analyzed based on Eqs.(4) and (5), respectively. (4) (5) where R is the number of all data points and yi is the value of ith data point. To convert multiple responses to single-response optimization via S/N ratio calculation, grey relational analysis was employed. The obtained results from Taguchi method is then calculated to determine the highest overall grey relational which represents the optimal parametric combination. Before grey relational analysis, data preprocessing is normally required, which is a process of transferring the original sequence to a comparable sequence that is normalized within the range of 0 to 1.The reference sequence and comparable sequence can be denoted by xo(k) and xi(k) for i = 1, 2, . . ., m; k = 1, 2, . . ., n, respectively, where m is the total number of 48

The 30th Special CU-af Seminar 2022 September 2, 2022 experiments to be considered and n is the total number of observation data. The appropriate equation for the normalization also depends on the type of the quality characteristic. In this work, the smaller-the-better and the-larger-the-better quality characteristics were applied for the normalization of all responses that as expressed in Eqs. (6) and (7), respectively. (6) (7) where xi(k) is the value after grey relational generation, minyi(k) is the smallest value of yi(k) for kth response, and maxyi(k) is the largest value of yi(k) for kth response. Grey relational coefficient can be calculated using Eq. (8). (8) where is the distinguishing coefficient, If all the process parameters have equal weighting, is set to be 0.5. The grey relational grade is the average of all grey relational coefficients, which can be determined using Eq. (9). (9) Finally, the optimal condition of sugarcane wax extraction can be the level corresponding to the highest value of average grey relational grade of each factor. 49

The 30th Special CU-af Seminar 2022 September 2, 2022 Results and Discussion Following DOE proposed by Taguchi’s method in terms of providing continuous length, randomly arranged, not curly, and no appearance of beads, it is necessary to note that the fiber with beads exhibited negative effect on the properties of the fiber mats[4]. Sixteen experiments were carried out from the conditions shown in Table 2. Figure 5 illustrates the morphology of PAN/bio-based PU electrospun showing the continuous fiber with no formation of beads with a diameter ranging from 260 to 1,160 nm. Influences of processing conditions, i.e., concentration of polymer solutions, applied voltage, and distance from tip to collector, on fiber diameter of PAN/bio-based PU electrospun fiber mats were investigated according to Taguchi technique. The fiber diameter sharply enlarged with increasing concentration of polymer solution, i.e., from 271 nm at 8 wt% concentration to 1,140 nm at 14 wt% concentration, which can be accounted to be 320% increment. This can be due to polymer in polymer solution being at higher concentration. As the polymer solution was ejected from the tip, the solvent evaporated, and polymer solidified into an enlarge fiber diameter at the collector. It was also attributed to the more viscous, higher surface tension and viscoelastic force of the solution[18]. The result was in accordance with that reported by Jacob et al.[19]. The electrolyte uptake of PAN/bio-based PU electrospun fiber mats was in the range of 1,650 to 1,971%, which is greater than 948% of the commercial glass fiber separator. The electrolyte uptake relates to the porosity of the membrane. Consequently, the discussion of electrolyte uptake is made in relation to porosity. Figures 6(a-c) show the relationship between electrolyte uptake and porosity with concentration, applied voltage, and distance from tip to collector. It can be observed that the electrolyte uptake of the samples increased with increasing porosity vice versa. Figure 5: Morphology of PAN/bio-based PU 50

The 30th Special CU-af Seminar 2022 September 2, 2022 Figure 6: Effect of (a) concentration, (b) applied voltage, and (c) distance from tip to collector on (circle) electrolyte uptake and (square) S/N ratio of PAN/bio-based PU at 75/25. The tensile strength of PAN/bio-based PU electrospun fiber mats were in the range of 20 to 44 MPa. Figures 7(a-c) show the relationship between tensile strength and fiber diameter with concentration, applied voltage, and distance from tip to collector. It is noticed in Figure 7 (a) that tensile strength of the samples showed a similar trend to fiber diameter, i.e., as the fiber diameter increased, the tensile strength increased. The tensile strength of our PAN/bio-based PU electrospun fiber mats were higher than the minimum requirement for separator. Figure 7: Effect of (a) concentration, (b) applied voltage, and (c) distance from tip to collector on (circle) tensile strength and (square) S/N ratio of PAN/bio-based PU electrospun fiber. Contact angle of separator indicates its wettability and ability to absorb electrolyte. Good wettability helps decrease the internal ionic resistance in the battery cell as well as retains electrolyte during cell assembly and prolong battery life under normal operation[20]. The contact angle measurements for the PAN/bio-based PU electrospun fiber mats are shown in Figure 8 indicating no significant effect of the electrospinning parameters. As can be seen, all electrospun fiber mats were hydrophilic nature and have a static contact angle in the range of 64.7–89.5°, which were less than 90°, whereas the neat PAN electrospun fiber mat showed the hydrophobic nature having the value of 116°, higher than 90°. The result indicated that the addition of bio-based PU could enhance hydrophilicity of the PAN/bio-based PU electrospun fiber mats, supporting high electrolyte uptake of the fiber mats, as the electrolyte used in this research was aqueous-based electrolyte and thus greater ionic conductivity. 51

The 30th Special CU-af Seminar 2022 September 2, 2022 Figure 8: Contact angle of PAN/bio-based PU membrane. The ionic conductivity of the electrolyte-immersed separator was determined from the bulk resistance value, i.e., the x-intercept of extrapolated Nyquist plot at high-frequency end of real Z. A separator with high porosity is generally characterized with sufficient ionic conductivity of a battery cell. The ionic conductivity of PAN/bio-based PU electrospun fiber mats was in the range of 1.08 to 3.11 mS/cm. Figures 9(a-c) show the relationship between ionic conductivity and porosity with concentration, applied voltage, and distance from tip to collector. It is observed that the ionic conductivity has a similar trend to porosity of the electrospun fiber mats, i.e., the ionic conductivity increased with porosity and vice versa. Higher porosity would enhance both electrolyte uptake rate as well as ionic conductivity; as a result, higher and faster ionic conductivity may be achieved, thus improving the overall performance of separator membrane[21, 22]. F i g u re 9 : E ff e c t o f ( a ) concentration, (b) applied voltage, and (c) distance from tip to collector on (circle) ionic conductivity and (square) S/N ratio of PAN/bio-based PU at 75/25. 52

The 30th Special CU-af Seminar 2022 September 2, 2022 Conclusions The bio-based PU was prepared using partially bio-based diisocyanate, polycaprolactone diol, and ethylene glycol at a mole ratio of 2.1:1:1. PAN/bio-based PU electrospun fiber mats were prepared using electrospinning technique. Among other processing parameters, the concentration of polymer was noted to have significantly affected the properties of the fiber mats. As per the results of optimization using grey relational analysis, the PAN/bio-based PU prepared with the processing condition of 14 wt% polymer concentration, 25 kV of applied voltage, and 16 cm of distance from tip to collector was the most suitable fabrication condition providing all best properties. It can be concluded that the PAN/bio-based PU electrospun fiber mats are a promising candidate for high flexibility and high-performance separator in Zn-ion battery. References 1. Xu W., Wang Y.: Recent progress on zinc-ion rechargeable batteries. Nano-Micro Letters, 11, 90 (2019). 2. Cao J., Zhang D., Zhang X., Sawangphruk M., Qin J., Liu R.: A universal and facile approach to suppress dendrite formation for a Zn and Li metal anode. Journal of Materials Chemistry A, 8, 9331-9344 (2020). 3. Wu L., Zhang Y., Shang P., Dong Y., Wu Z. S.: Redistributing Zn ion flux by bifunctional graphitic carbon nitride nanosheets for dendrite-free zinc metal anodesk. Journal of Materials Chemistry A, 9, 27408-27414 (2021). 4. Ma X., Kolla P., Yang R., Wang Z., Zhao Y., Smirnova A. L., Fong H.: Electrospun polyacrylonitrile nanofibrous membranes with varied fiber diameters and different membrane porosities as lithium-ion battery separators. Electrochimica Acta, 236, 417-423 (2017). 5. Belov, D., and D.-T. Shieh, GBL-based electrolyte for Li-ion battery: thermal and electrochemical performance. Journal of Solid State Electrochemistry, 16(2), 603-615 (2011). 6. Ghosh M., Vijayakumar V., Kurungot S.: Dendrite growth suppression by Zn2+-integrated Nafion ionomer membranes: beyond porous separators toward aqueous Zn/V2O5 batteries with extended cycle life. Energy Technology, 7, 1900442 (2019). 7. Lee, H., Yanimaz, M., Torakci, O., Fu, K., and Zhang, X., A review of recent developments in membrane separators for rechargeable lithium-ion batteries. Energy Environmental and Science, 7(12), 3857-3886 (2014). 8. Babar, A., Wang, X., Yu, J., and Ding, Bin., Introduction and historical overview, in Electrospinning Nanofabrication and Applications, X.W. Bin Ding, Jianyong Yu, Editor. 2019, ELSEVIER: Amsterdam, Netherlands. 9. Wu, N., Jing, B., Wang, X., Kuang, H., and Wang, Q., A novel electrospun TPU/PVdF porous fibrous polymer electrolyte for lithium ion batteries. Journal of Applied Polymer Science, 125(4), 2556-2563 (2012). 10. Kim, B.G., Kim. J.S., Min, J., Lee, Y.H., Choi, J.H., Jang, M.C., Freunberger, S.A., Choi, J.H., and Choi, J.W., A Moisture- and Oxygen-Impermeable Separator for Aprotic Li-O2Batteries. Advanced Functional Materials, 26(11), 1747-1756 (2016). 11. Zhou, L., Cao, Q., Jing, B., Wang, X., Tang, X., and Wu, N., Study of a novel porous gel polymer electrolyte based on thermoplastic polyurethane/poly(vinylidene fluoride-co- hexafluoropropylene) by electrospinning technique. Journal of Power Sources, 263, 118-124, (2014). 53

The 30th Special CU-af Seminar 2022 September 2, 2022 12. Liu X., Song K., Lu C., Huang Y., Duan X., Li S., Ding Y. H.: Electrospun PU@GO separators for advanced lithium ion batteries, Journal of Membrane Science, 555, (2018). 13. Kiai M.S., Kizil H., Electrospun nanofiber polyacrylonitrile coated separators to suppress the shuttle effect for long life lithium-sulfur battery. Journal of Applied Polymer Science, 137. 48606 (2019). 14. Gopalan, A., Santhosh P., Manesh K., Nho J. Kim S. Hwang C. Lee K., Development of electrospun PVdF–PAN membrane-based polymer electrolytes for lithium batteries. Journal of Membrane Science, 325(2), 683-690 (2008). 15. Chen D., Wang X., Liang J., Zhang Z., Chen W.: A novel electrospinning polyacrylonitrile separator with dip-coating of zeolite and phenoxy resin for Li-ion batteries. Membranes, 11, (2021). 16. Elia G. A., Ducros J. B., Sotta D., Delhorbe V., Brun A., Marquardt K., Hahn R.: Polyacrylonitrile separator for high-performance aluminum batteries with improved interface stability. ACS Applied Materials & Interfaces, 9, 38381-38389 (2017). 17. Roy R. K.: A primer on the Taguchi method. Society of Manufacturing Engineers, New York, (1990). 18. Mi H. Y., Jing X., Napiwocki B. N., Hagerty B. S., Chen G., Turng L. S.: Biocompatible, degradable thermoplastic polyurethane based on polycaprolactone-block- polytetrahydrofuran-block-polycaprolactone copolymers for soft tissue engineering. Journal of Materials Chemistry B, 5, 4137-4151 (2017). 19. Jacobs V., Anandjiwala R.D., Maaza M.: The influence of electrospinning parameters on the structural morphology and diameter of electrospun nanofibers. Journal of Applied Polymer Science, 115, 3130-3136 (2010). 20. Tipduangta P., Sirithunyalug J.: Fundamental and application of electrospinning technology in pharmaceuticals and cosmetics. Isan Journal of Pharmaceutical Sciences, 13, 1-15 (2017). 21. Choi W., Shin H. C., Kim J. M., Choi J. Y., Yoon W. S.: Modeling and applications of electrochemical impedance spectroscopy (EIS) for lithium-ion batteries. Journal of Electrochemical Science and Technology, 11, 1-13 (2020). 22. Liu X., Song K., Lu C., Huang Y., Duan X., Li S., Ding Y. H.: Electrospun PU@GO separators for advanced lithium ion batteries, Journal of Membrane Science, 555, (2018). 54



Label-Free Identification and Classification of Circulating Tumor Cells using Deep Learning and High-Content Imaging Sira SRISWASDI

The 30th Special CU-af Seminar 2022 September 2, 2022 Label-Free Identification and Classification of Circulating Tumor Cells using Deep Learning and High-Content Imaging Sira SRISWASDI1* Abstract Circulating tumor cells (CTC) are shed from primary tumor into the patient’s bloodstream and can initiate metastasis in other parts of the body. Existing methods for detecting CTC rely on size filtering and specific cell surface markers which are biased toward certain cancer and cell types. Recent advances in deep learning have shown that artificial neural network models can be trained to accurately classify cell types and distinguish various cell compartments in unlabeled bright-field microscopy images. Several studies have successfully applied this technique to differentiate between spiked-in cancer cells and normal blood cells in unlabeled microscopy images. However, their evaluations were based on cell line-derived cancer cells whose morphology is homogeneous and capture only a narrow population of CTC in patient blood. Without exposing CTC classification models to the morphological heterogeneity of cancer and normal cells from diverse tissues, the models would likely confuse normal non-blood cells with CTC. In this research, we (i) constructed a database of microscopic images of cancer and normal cell with diverse morphology and (ii) developed a proof-of-concept deep neural network model that can distinguish between cancer and normal cells derived from the same tissue and patient. Cancer and adjacent normal tissues were biopsied from 3 cholangiocarcinoma patients and 3 colorectal cancer patients to grow 3D organoids. Cells from cancer and normal organoid were labeled with different fluorescence dyes, mixed, and subject to high-content microscopic imaging. In total, more than 3000 paired bright-field and fluorescence microscopic images, each containing more than 40-50 cells, were acquired to establish a realistic database of cell images that capture the heterogeneous morphology of cancer cells. A two-stage deep neural network model was developed to distinguish between cancer and normal cells derived from the same tissue with a precision of 0.72, a recall of 0.58, and an area under the receiver operating characteristics curve (AUROC) of 0.82. The model was also robust to patient-to-patient variation as it performed well on cell images from an unseen patient with less than 0.03 drop in performance. Our research serves as a foundation for an automated, universal CTC detection platform that can recognize normal non-blood cells and determine the tissue-of-origin for each detected CTC. 1Research Affairs Faculty of Medicine, Chulalongkorn University Bangkok, Thailand 57

The 30th Special CU-af Seminar 2022 September 2, 2022 Introduction and Objectives Over the past few decades, lithium ion batteries (LIBs) have attracted considerable attenCirculating tumor cell (CTC), or cell from primary tumor that were shed into the patient’s bloodstream, holds important clinical values as a source of early, non-invasive biomarker of metastasis and cancer prognosis and many cancer types[1,2]. Existing technologies for isolating and detecting CTC mainly rely on the fact that most normal blood cells can be captured by antibody targeting certain cell surface markers, such as CD45, while tumor cells can be captured by antibody targeting different markers[3]. Although multiple antibodies have been developed for characterizing various CTC types, such as epithelial and mesenchymal CTC[4], enrichment-based approaches still cannot account for the full heterogenicity of CTC. In fact, a study of lung cancer patients has shown that only 40-60% of CTC in blood samples were detected by enrichment-based approaches[5]. Nowadays, high-throughput sequencing technologies have also been applied to characterize the genome and transcriptome of individual CTC[6] as a non-invasive mean to probe the molecular signature of primary tumors and to develop prognostic cancer biomarkers. Another possibility for unbiased characterization of individual CTC is through high-content microscopy imaging of patient blood samples, whereby cancer cells can be differentiated from normal cells as well as classified into types based on their distinctive morphological properties[7, 8]. These techniques are enabled by recent advances in deep learning which let us train artificial neural network models to accurately identify cell types[9,10] and pinpointing the locations of subcellular compartments[11, 12] from bright-field microscopy images without any labeling of the cells. However, imaging-based CTC detections were mostly developed and/or validated only on spiked-in cells from a few cell lines that do not capture the broad heterogeneity and morphological properties of actual CTC[13]. For example, [14] trained a deep learning model using 436 cultured cells and 1,309 white blood cells and validated their model on 32 CTCs from two patients. A recent work has also shown that CTCs derived from different tumor sites exhibit clearly distinct morphological characteristics[15]. This finding also suggests the possibility of developing an imaging-based CTC detection platform that can also predict the tissue-of-origin of each detected CTC. Hence, the first step toward developing a generalized imaging-based CTC detection platform is to establish a large-scale microscopy imaging database of cancer and normal cells that capture the heterogeneity of both cancer types and tissue types. Patient-derived organoids, or 3D cultures, have been shown as realistic sources of diverse cell types and morphology that faithfully represent the genotype and phenotype of cancer subtypes[16, 17]. The combination of paired cancer and normal cells derived from the same tissue of the same patient would serve as a good benchmark for an imaging-based CTC detection technique by testing whether the technique can distinguish between cancer and normal cells (as supposed to distinguishing between blood and non-blood cells). By expanding the database of cell images to cover multiple tissues, cancer types, and patients, and by linking cell images to prognosis and treatment response information, future imaging-based CTC platforms have the potential to not only detect CTC, but also predict the tissue-of-origin and aid clinical decision making. In this research, a large database of microscopic images of more than 100,000 individual organoid-derived cancer and normal cells from 3 cholangiocarcinoma patients and 3 colorectal cancer patients were constructed, and a proof-of-concept deep neural network model was developed to (i) evaluate the possibility of distinguishing cancer and normal cells based on only unlabeled bright-field microscopic images and (ii) explore the morphological diversity of cancer and normal cells across cancer types and individual patients. The full dataset and developed deep 58

The 30th Special CU-af Seminar 2022 September 2, 2022 learning model will be made publicly available on FigShare and GitHub. Methods Cholangiocyte organoid culture Liver biopsies were cut into small pieces and washed 3 times with Advanced DMEM/ F12 supplemented with 1x Glutamax, 10 mM HEPES, and 1x antibiotics (AdDF+++, Gibco, Thermo Scientific). Liver tissues were digested using 100 μg/ml dispase I and 300 U/ml collagenase XI in Cholangiocyte culture media with Advanced DMEM/F12 containing 10% R-Spondin condition media, 10% Wnt3a condition media, 1 mM N-Acetylcysteine, 10 mM Nicotinamide, 1x B27 supplement, 1x N2 supplement, 100 ng/ml Noggin, 10 nM Gastrin-I, 50 ng/ml EGF, 5 uM A83-01, 100 ng/ml FGF10 (Peprotech), 25 ng/ml HGF (R&D Systems), and 10 μM FSK (Tocris). The cultures were incubated at 37 °C for 1 hour. The digestion reaction was stopped with 10 ml AdDF+++ and the resulting suspension was filtered through a 70 μM cell strainer. Cells in suspension were collected via centrifugation and washed 5 times with AdDF+++. Cell pellets were resuspended in 70% Matrigel (Corning) and dropped on pre-warmed 24-well culture plates. After the Matrigel solidified, 500 μl of organoid culture media was added. Cells were cultured at 37 °C with 5% CO2. The media were changed every 3 days and the cell passage was performed every 1-2 weeks by mechanically dissociating the cells with P1000 pipette tip. Colorectal organoid culture Colorectal cancer and normal adjacent tissues were biopsied, cut into 1x1 mm pieces, and washed 5 times with AdDF+++. Colorectal tissues were digested using dispase II and collagenase D (Roche, Sigma Aldrich) in Colorectal culture media with Advanced DMEM/F12 containing 10% R-Spondin condition media, 10% Wnt3a condition media, 1 mM N-Acetylcysteine, 10 mM Nicotinamide, 1x B27 supplement, 1x N2 supplement, 100 ng/ml Noggin, 1 μM Gastrin-I, 50 ng/ml EGF, 50 uM A83-01, and 1 μM SB202190. The cultures were incubated at 37 °C for 1 hour. The digestion reaction was stopped with 10 ml AdDF+++ and the resulting suspension was filtered through a 70 μM cell strainer. Cells in suspension were collected via centrifugation and washed 5 times with AdDF+++. Cell pellets were resuspended in 70% Matrigel (Corning) and dropped on pre-warmed 24-well culture plates. After the Matrigel solidified, 500 μl of organoid culture media was added. Cells were cultured at 37 °C with 5% CO2. The media were changed every 4 days and the cell passage was performed every 1-2 weeks by mechanically dissociating the cells with P1000 pipette tip. Fluorescence labeling and high-content imaging Each organoid was dissociated into single cells using TrypLETM Express Enzyme (Gibco, Thermo Scientific). Around 106 cancer and normal cells were obtained from each sample. Cells from cancer organoids were stained with a deep red fluorescence (Cytopainter ab176736) while cells from normal organoids were stained with green fluorescence (Cytopainter ab176735). Nuclei were stained with Hoechst. Cancer and normal cells were mixed at 1:1 ratio, dropped on 96-well plates, and subjected to bright-field and fluorescence imaging on an Opera PhenixTM instrument (Perkin Elmer). In total, 1632 paired bright-field and fluorescence images were acquired for cancer and normal cholangiocytes. 1851 images were acquired for colorectal cells. Each image consists of 1080x1080 pixels and contains 20-30 individual cells on average. 59

The 30th Special CU-af Seminar 2022 September 2, 2022 Image processing and dataset preparation Following prior protocol[11], all fluorescence images were subjected to 5x5 median filtering until convergence to reduce salt-and-pepper noises. Both bright-field and fluorescence images were than down-sampled bilinearly by a factor of two to reduce shot noises. Finally, pixel intensities were normalized per image to the same mean and standard deviation. In contrast to prior study, we found that flat field correction and dust artifact removal were not necessary for improving the quality of images here. To aid in the annotation of cancer and normal cells in each bright-field image, the locations of every cell in a small set of 30 images were manually annotated with bounding boxes by a scientist with >2 years of experience in microscopic imaging. A deep neural network model based on the Faster-RCNN architecture[18] was then trained on these images and used to predict cell locations and preliminary cancer/normal/uncertain labels for all cells in all other images. Cell annotation Three annotators were recruited. One annotator is an expert in microscopy with more than two years of experience. The other two annotators are biology master’s degree students with some cell biology and microscopy experiences. Inter-annotator agreement was evaluated at the beginning of the research by asking all three annotators to analyze the same set of 6 images (about 150 individual cells). It should be noted that all annotators received assistance from the deep neural network model which provided preliminary cell locations and classification labels. Once an acceptable to agreement was confirmed, each annotator was asked to analyze different images to maximize the coverage of the dataset. The LabelMe[19] tool was used for the annotation process. Annotators were allowed to add/remove cells and change classification labels with full information from both bright-field and fluorescence images. The main purpose of the manual annotation process was to remove noises and cell debris from the dataset. Currently, 1087 out of 1632 images of cholangiocytes had been annotated by at least one annotator. Additionally, a test set of 120 images were randomly selected for which two annotators were asked to annotate cancer cells without the help of the deep neural network model. The annotations were merged using the non-maximum suppression technique. When the two annotators produced slightly overlapped cell locations, the annotation made by the more experienced one was selected. Model development A two-stage deep neural network model architecture, which has been applied to similar tasks of detecting small objects in large images[20, 21], was used here. The Faster R-CNN network[18] with ResNet-50[22] as the backbone was used as the detector. Another deep convolutional neural network with EfficientNet-B4[23] as the backbone was used as the classifier. Both networks were initialized with ImageNet[24] pretrained weights. Minor modifications were made to the networks to make it compatible with our input image resolution and the number of input channels (The RGB channels used in ImageNet data were replaced by bright-field and fluorescence channels). The flow of the two-stage architecture is as follows. First, the detector receives a 1080x1080 bright-field image as input and proposes regions in the image where cells are located along with classification confidences of whether each cell is cancer or normal. Each proposed cell regions are then resized to 128x128 and fed into the classifier as input to obtain the final classification confidences. In some cases, the classification confidences made by the detector, Sdet, and the classifier, Scls, can be weighted-average to improve the performance according to the formula Sfinal = ω Sdet + (1 – ω) Scls. 60

The 30th Special CU-af Seminar 2022 September 2, 2022 The detector was trained according to an earlier object detection framework[25] with Stochastic Gradient Descent (SGD) as the optimizer, a batch size of 8 images, and a learning rate schedule over 8 epochs which starts at 10-3 and is reduced by 10-fold after 5 and 7 epochs have passed. Random flip and brightness augmentations were performed on the input. The classifier was trained with a classification framework with Adam as the optimizer, a batch size of 64, and a learning rate schedule over 12,000 epochs which starts at 5x10-4 and is reduced by 10-fold after 6,000 and 10,000 epochs have passed. Random geometric transformation and brightness augmentations were performed on the input. Performance evaluations Because the detection of cancer cell is the primary goal of this research, model performances were measured based on the ability of the model to detect cancer cells. Precision, or positive predictive value, recall, or sensitivity, F1, which is the harmonic mean of precision and recall, and area under the receiver operating characteristics curve (AUROC) were reported. In each analysis, the confidence cutoff threshold that yielded the highest F1 score was selected for calculating the scores. In addition to the cutoff-based precision score, average precision, which roughly corresponds to the area under the precision-recall curve and does not depend on the choice of confidence cutoff threshold, was also calculated. A prediction is considered correct if the predicted bounding box overlaps the ground truth bounding box with at least 0.5 intersection-over-union score (IOU) and the predicted class matched the ground truth label. An IOU score between two bounding boxes is calculated by measuring the area of the overlap between the two boxes and dividing that by the area of the union of the two boxes. Results and Discussion In total, 1632 paired bright-field and fluorescence images were acquired for cancer and normal cholangiocytes, and 1851 paired images were acquired for colorectal cells. As shown in Figure 1, each image consists of 1080x1080 pixels and contains at least 20-30 individual cells on average. Using the paired fluorescence images, each cell can be labeled as cancer (based on red signal), normal (based on green signal), or unknown (if there is no clear red or green signal). To aid the annotation of cell locations and classes, a small set of 30 images were manually annotated by an expert and the resulting annotations were used to train a Faster R-CNN model (Figure 2), a state-of-the-art neural network that can efficiently identify small individual objects in a much larger image[18]. The model was then used to provide preliminary predictions for cell locations and classes that will be revised by human annotators (Figure 3). This scheme greatly reduced the amount of workload for the human annotators because the most time-consuming part is drawing the bounding boxes of individual cells. Nonetheless, even with assistance from the neural network model, full annotation of each image still required up to 10-15 minutes by a human annotator. To ensure high annotation quality, multiple annotators with different experiences with cell microscopy were asked to analyze the same images. As comparison of their annotations show a good agreement (Figure 4), we decided to assign each annotator to analyze a different set of images to maximize the coverage of the dataset. However, it should be noted that the amount of experience that the human annotator has does have some impact on the development of neural networks for distinguishing cancer and normal cells. On the test set of 120 images, the final model’s predictions agree more with the annotations made by the human annotator with more experience than with other annotators. But the difference is quite small (0.023 higher average 61

The 30th Special CU-af Seminar 2022 September 2, 2022 precision when evaluated on the annotations made by annotator with more experience). Currently, 1087 out of 1632 images of cholangiocytes had been fully annotated. This corresponds to a total of 32834 individual cancer cells, 25623 individual normal cells, and 27710 individual unknown cells. Figure 1: Examples of the bright-field and fluorescence images acquired in this research. The left panel shows the full 1080x1080 pixel bright-field image that contains both cancer and normal cells. The right panel show individual cancer cells (red, top row), normal cell (green, middle row), and unknown cells (blue, bottom row) which display only the Hoechst staining. Figure 2: Faster R-CNN architecture[18]. From bottom to top, the first convolutional layers (conv layers) extracted image-based features from the large input image. The features were first used to identify foreground objects by the Region Proposal Network and later pooled with the proposed image patches to classify individual objects. This network design is efficient because it reuses features extracted by the early convolutional layers in all downstream modules. It should be noted that even without human annotation, the combination of bright-field and fluorescence images were enough to train a deep neural network to accurately determine cell locations and boundaries (Figure 5). Hence, the primary contributions of human annotators are to remove non-cell objects, such as dead cells and cell debris, and to revise the cancer/normal/ unknown classification labels. 62

The 30th Special CU-af Seminar 2022 September 2, 2022 Figure 3: Development scheme for the preliminary deep neural network model for aiding cell location and class annotation process by human annotators. The first step involved manual annotations on 30 images by an expert. This data was then used to train a Faster R-CNN model, which was then used to generate preliminary predictions of the locations and classes of cells in all images. These automated annotations were then used as pseudo labels to further refine the initial Faster R-CNN model. Figure 4: Comparison of cancer cell annotations made by two human annotators (red and green bounding boxes). The first annotator (red) has more than 2 years of experience with cell microscopy while the second annotator (green) is a biology master’s degree student with some experience in cell biology and microscopy. Both annotators mostly agree on the same bounding boxes. 63

The 30th Special CU-af Seminar 2022 September 2, 2022 Figure 5: Ability of neural networks to identify cell boundaries without the aid of human annotation. In this setting, the model was trained to predict the fluorescence image from an input bright-field image. Cell boundaries (mask generated by the model) were defined as pixels with high predicted fluorescence signals. The predicted cell boundaries can be used to denoise fluorescence images by removing fluorescence signals that bled into the background pixels. The task of detecting cancer cells was divided into two stages: detection stage and classification stage (Figure 6). This two-stage approach has been successfully applied in prior research that aimed to identify small objects in a much larger input image[20, 21]. During the detection stage, a trained object detection network such as Faster R-CNN processes the input image to identify potential cell locations, in the form of bounding boxes, and simultaneously produce preliminary classifications. The image regions that correspond to the predicted bounding boxes were then extracted as input into the classification stage, which consists of a convolutional neural network (CNN) model. An EfficientNet-B4 architecture[23] was used as the classifier here because it requires less resources and is more efficient than most other CNN architectures. Even though the detector already performed classification, the additional classification stage is needed because the classification confidence scores produced by the detector are often poor. A dedicated classification network in the classification stage helps refine these confidence scores to better match the ground truth labels. As shown in Figure 6, the final classification output can significantly differ from the initial proposal made by the detection stage. To train this network, data from 947 annotated bright-field images (more than 23000 cells from each class) were used, 64

The 30th Special CU-af Seminar 2022 September 2, 2022 and the remaining annotated images were used for validation and testing. Figure 6: The diagram of a two-stage detection-classification framework. In the detection stage, an object detection network proposes cell locations and classes based on the input bright-field image. In the classification stage, the proposed regions were extracted and fed into another classification network. The final classification outputs were then post-processed and overlayed on top of the input bright-field image. The two-stage network achieved 0.54 F1 and 0.77 AUROC with bright-field images as input (Table 1). Inclusion of the blue fluorescence signals which correspond to the nuclei and can realistically be obtained in real applications further boosted the performance to 0.65 F1 and 0.82 AUROC. This improvement likely resulted from the fact that the Hoechst staining can capture the difference in morphology, size, and DNA content of the nuclei between cancer cells and normal cells[26, 27]. One interesting result is how information from unknown cells (those with no clear red and green fluorescence signal) should be used to further improve the two-stage cancer cell detection network. As shown in Table 2, not including unknown cells in the training data at all yielded the worst outcome. Predicting pseudo labels for the unknown cells and using them for training helped the model learns more about cell diversity (+0.036 average precision compared to not using unknown cells). Including only high-confidence pseudo labels yielded a better performance than believing all pseudo labels (+0.085 average precision compared to not using unknown cells). Curiously, the best improvement (+0.106 average precision compared to not using unknown cells) was achieved if all unknown cells were treated as normal. This is unexpected because when the latent representation of each input image was extracted from the output of the last global average pooling layer of the model and visualized on 2-dimension using the Uniform ManifoldApproximation and Projection (UMAP) technique, many unknown cells were placed together with cancer cells (Figure 7). This distribution pattern strongly suggests that many unknown cells were indeed cancer cells that might be poorly stained. However, because most unknown cells were placed in a separate region from both cancer and normal cells (Figure 7, middle right region of the UMAP space), it is possible that treating all 65

The 30th Special CU-af Seminar 2022 September 2, 2022 unknown cells as normal (i.e., non-cancer) helps the model better delineate the morphological boundary of cancer cells. Indeed, a slight cancer cell detection improvement was obtained by first clustering cells based on the latent representations derived from the neural network model and assigning each unknown cell as cancer, normal, or unknown based on the majority class of the cluster it belongs to. Figure 7: 2D UMAP visualization for the latent representation of individual cells derived from the output of the last global average pooling of the two-stage neural network. These latent representations are essentially image-based features that the model used for classifying cells. Red indicates cancer cells. Green indicates normal cells. Blue indicates unknown cells. The left UMAP plot was derived from the model variant trained on only cancer and normal cells. The right UMAP plot was derived from the model variant trained on cancer, normal, and unknown cells. Unknown cells were assigned as cancer, normal, or remained unknown based on a k-mean clustering of latent representations of individual cells. Lastly, we evaluated the extent of patient-to-patient variation in cell morphology by training the two-stage model using data from one or two patient(s) and calculating the performance on data from the unseen patient(s). Similar performances were achieved (less than 0.02 difference in average precision) when the model was trained and tested on cell images derived from different cholangiocarcinoma patients (Table 3). Although the performances were lowest when trained or tested on the 3rd cholangiocarcinoma patient, this is more likely because only 21 images from this patient have been annotated by human experts. In contrast, around 500 images were annotated each for the 1st and 2nd patients. 66

The 30th Special CU-af Seminar 2022 September 2, 2022 Conclusion In this research, we have established a database of more than 3,000 bright-field microscopic images of organoid-derived cancer and normal cells from 2 tissues and 6 patients. Paired fluorescence images allowed the detailed annotation of individual cells both automatically by an object detection network and manually by human experts. In total, more than 75,000 images of individual cells were annotated. This enabled the development of a proof- of-concept two-stage neural network that can detect cancer cells from an unlabeled bright-field microscopic image with a precision of 0.75, a recall of 0.49, and an AUROC of 0.77. Inclusion of nuclei fluorescence staining, which can be practically performed in actual applications, further improve these performances to a precision of 0.72, a recall of 0.58, and an AUROC of 0.82. The minimal impact from patient-to-patient variation on the model performance suggested that our model can potentially generalize to data from future patients. For the next steps, we plan to complete the annotations for both cholangiocarcinoma and colorectal cancer datasets and explore the model’s behaviors across cancer types. Organoid-derived cells from other cancer types will also be acquired to expand the database. References 1. S. Rawal, Y. P. Yang, R. Cote, A. Agarwal, Identification and Quantitation of Circulating Tumor Cells. Annu Rev Anal Chem (Palo Alto Calif) 10, 321-343 (2017). 2. Y. Ming et al., Circulating Tumor Cells: From Theory to Nanotechnology-Based Detection. Front Pharmacol 8, 35 (2017). 3. P. Bankó et al., Technologies for circulating tumor cell separation from whole blood. J Hematol Oncol 12, 48 (2019). 4. A. Satelli, Z. Brownlee, A. Mitra, Q. H. Meng, S. Li, Circulating tumor cell enumeration with a combination of epithelial cell adhesion molecule- and cell-surface vimentin-based methods for monitoring breast cancer therapeutic response. Clin Chem 61, 259-266 (2015). 5. Y. Xu et al., Circulating tumor cell detection: A direct comparison between negative and unbiased enrichment in lung cancer. Oncol Lett 13, 4882-4886 (2017). 6. Z. Zhu, S. Qiu, K. Shao, Y. Hou, Progress and challenges of sequencing and analyzing circulating tumor cells. Cell Biol Toxicol 34, 405-415 (2018). 7. A. Ciurte, C. Selicean, O. Soritau, R. Buiga, Automatic detection of circulating tumor cells in darkfield microscopic images of unstained blood using boosting techniques. PLoS One 13, e0208385 (2018). 8. C. Aguilar-Avelar et al., High-Throughput Automated Microscopy of Circulating Tumor Cells. Sci Rep 9, 13766 (2019). 67

The 30th Special CU-af Seminar 2022 September 2, 2022 9. K. Yao, N. D. Rochman, S. X. Sun, Cell Type Classification and Unsupervised Morphological Phenotyping From Low-Resolution Images Using Deep Learning. Sci Rep 9, 13467 (2019). 10. C. L. Chen et al., Deep Learning in Label-free Cell Classification. Sci Rep 6, 21471 (2016). 11. E. M. Christiansen et al., In Silico Labeling: Predicting Fluorescent Labels in Unlabeled Images. Cell 173, 792-803.e719 (2018). 12. R. Brent, L. Boucheron, Deep learning to predict microscope images. Nat Methods 15, 868-870 (2018). 13. S. Park et al., Morphological differences between circulating tumor cells from prostate cancer patients and cultured prostate cancer cells. PLoS One 9, e85264 (2014). 14. S. Wang et al., Label-free detection of rare circulating tumor cells by image analysis and machine learning. Sci Rep 10, 12226 (2020). 15. L. Zeune et al. (Nature Machine Intelligence, 2020), vol. 2, pp. 124-133. 16. J. Drost, H. Clevers, Organoids in cancer research. Nat Rev Cancer 18, 407-418 (2018). 17. F. Amato, C. Rae, M. G. Prete, C. Braconi, Cholangiocarcinoma Disease Modelling Through Patients Derived Organoids. Cells 9, (2020). 18. S. Ren, K. He, R. Girshick, J. Sun, Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. IEEE Trans Pattern Anal Mach Intell 39, 1137-1149 (2017). 19. B. Russell, A. Torralba, K. Murphy, W. Freeman. (International Journal of Computer Vision, 2008), vol. 77, pp. 157-173. 20. C. Li, X. Wang, W. Liu, L. J. Latecki, DeepMitosis: Mitosis detection via deep detection, verification and segmentation networks. Med Image Anal 45, 121-133 (2018). 21. H. Takimoto, Y. Sato, A. Nagano, K. Shimizu, A. Kanagawa. (Ecological Informatics, 2021), vol. 66, pp. 101466. 22. K. He, X. Zhang, S. Ren, J. Sun, in Proceedings of the IEEE conference on computer vision and pattern recognition. (2016), pp. 770-778. 23. M. Tan, Q. Le. (Proceedings of the 36th International Conference on Machine Learning, 2019). 24. J. Deng et al. (IEEE Conference on Computer Vision and Pattern Recognition, 2009). 25. K. Chen et al. (arXiv, 2019), pp. arXiv:1906.07155. 26. D. Zink, A. H. Fischer, J. A. Nickerson, Nuclear structure in cancer cells. Nat Rev Cancer 4, 677-687 (2004). 27. P. Jevtić, L. J. Edens, L. D. Vuković, D. L. Levy, Sizing and shaping the nucleus: mechanisms and significance. Curr Opin Cell Biol 28, 16-27 (2014). 68



Enhancing Local Capability toward Sustainable Municipal Solid Waste Management: Case Study of Nan Municipality, Thailand Wilailuk NIYOMMANEERAT Ph.D., Vacharaporn SOONSIN Ph.D., Puntita TANWATTANA Ph.D.

The 30th Special CU-af Seminar 2022 September 2, 2022 Enhancing Local Capability toward Sustainable Municipal Solid Waste Management: Case Study of Nan Municipality, Thailand Wilailuk NIYOMMANEERAT Ph.D.1*, Vacharaporn SOONSIN Ph.D.1*, Puntita TANWATTANA Ph.D.1* Abstract The research focuses on the development of the sustainable municipal waste management system in Nan Municipality and to enhance capacity building of the community and the government agencies. Nan walking street was selected as case study involved various stakeholder such as community, local government, and tourist sector. It suffered from overtourism and large amount waste generation with concerned from local authorities. Our results show that the most waste generated in Nan walking street are general waste (51%) followed by food waste (33%) bagasse containers, plastic bottles, wooden skewer, and others (2-7%). The study presents indicators for project evaluation in three aspect as social, environmental, and economic benefit from the project. The SROI as economic benefit showed two activities are worth for investment. Environmental benefit from plastic waste reduction that can be calculated for reduction of GHGs emission. Social benefit show increase of household income, and job creation from the project. 1Environmental Research Institute, Chulalongkorn University 71

The 30th Special CU-af Seminar 2022 September 2, 2022 Introduction and Objectives Several strategic approaches have been implemented since municipal solid waste is ranked as a significant issue in Thailand, which needs to be managed properly, otherwise there will be land-space constraints and at the end resulting in illegal dumping. Moreover, improper handling or disposal of municipal solid waste can cause severe environmental and health impacts by releasing pollutants into the landfill’s leachate and possibly potentially migrate out of the landfill site or transport to groundwater leading to groundwater contamination. Therefore, waste management capacity building should be considered as one of the important methodologies for sustainable solid waste management in Thailand. Additionally emergent environmental concerns and increasing waste management initiatives are the main factors powering the demand for waste recycling and waste composting improvements. Also, public awareness of waste management and monitoring planning are the strategics for enhancing the capacity of local authorities in sustainable solid waste management. Consequently, this research project has been proposed to focus on acquisition knowledge in science and social science in order to develop sustainable municipal waste management approach in Nan province, Thailand especially from tourism sectors to reduce the environmental risk from improper disposal of municipal solid waste. For this project, we have chosen Nan Municipality, located in the northern part of Thailand as a case study. Nan province has been selected to be the No.1 ASEAN Clean Tourist City Standard in 2018. Currently Nan province is one the tourist attractive destination for many visitors from all over the world and travelers from Thailand. Therefore, Nan’s economics is getting to increase radically to serve the consumption demand including hotels, restaurants, cafés/coffee shops and tourism activities. The amount of waste generated in Nan municipality is about 30 tons per day. When combine with other municipalities, there will be approximately 70 tons of waste per day. All municipal wastes are sorted at the waste disposal facility. Most of the waste is organic waste which accounts for more than 48 percent (food waste, and yard waste) and recyclable waste, accounting for 18 percent of the total waste. Based on our survey in Nan municipality in 2019, the waste that has not been properly sorted, utilized, and recycled were sent to dispose in landfills. Consequently, an unhygienic environment around the landfill area including the increase of population and tourists in Nan province every year which is likely to cause an increase in waste generated to landfills. Hence, the proper waste management is important for Nan municipality. The mismanaged waste can affect the environment and consequence to human health because of the limitation of the landfill areas in Nan province. Together with the Mayor of Nan Municipality intends to reduce the amount of wastes to be zero to landfill and realized that the waste can be utilized as a valuable resource, recycled, and sell as community income. This research project is mainly focusing on the development of the sustainable municipal solid waste management system in Nan Municipality and to enhance capacity building of the community and the government agencies related to municipal solid waste management. As a result, there will be sustainable municipal solid waste management models in Nan province. Project Contribution to SDGs Goals 72

The 30th Special CU-af Seminar 2022 September 2, 2022 Objectives 1. To enhance local capacity on waste management 2. To characterize waste composition and to investigate waste generation rate from various types of stakeholders 3. To promote sustainable waste management by increasing the amount of recycled and composted municipal solid waste 4. To propose the sustainable waste management and policy recommendation to municipality Methods Nan Walking Street (Kad Kuang Mueng) located in Nan province; Thailand was selected as representative case study (Figure 1.) Figure 1: Location of Nan’s walking street, Nan Municipality, Thailand Adapted from https://palanla.com/index.php?op=domesticLocation-detail&id=780 Nan Walking Street is one of the most visited destinations in Nan province. From our preliminary survey in 2020, the highest visitor were 3000 persons per day which generated waste more than 1000 kg per day. In effect, tourist destinations suffered from overtourism and cause the large accumulation of waste. Nan municipality mayor concerned on large generation 73

The 30th Special CU-af Seminar 2022 September 2, 2022 of waste to landfill which will reach capacity in less than 5 yrs. Otherwise, the mismanaged waste can affect the environment and consequence to exceed the capacity of the landfill areas in Nan province. Together with the Mayor of Nan Municipality intends to reduce the amount of wastes to be zero to landfill and realized that the waste can be utilized as a valuable resource, recycled, and sell as community income. The study aimed to enhance local capability toward sustainable municipal solid waste management in Nan municipality. This is action research through participatory process involved in various key stakeholders in waste management practice such as local authorities, local entrepreneurs, tourists, and Nan’s citizens in Nan municipality area. The details activities and project implementing are shown below. 1. Determining the stakeholders of waste management system and existing waste management practice in Nan’s walking street 2. Investigating and collecting baseline data of waste generation rates and composition in Nan’s walking street 3. Communicating and dissemination on waste management practice, implementing activities on waste reduction mechanism designs/campaign to enhance local capability on waste management practice in Nan’s walking street 4. Evaluation of projects achievement in Nan’s walking street (Figure 2). It is divided into two sub-projects as follows (1) Bamboo products replace plastic items such as bags, container and cup holder which produced from community enterprise within Nan province (2) Waste segregation at source and install separated garbage bins Figure 2: Indicators for project evaluation Social return on investment (SROI) is a method of accounting for the social, economic, and environmental value created by project/company. The purpose of issuing SROI is for corporations/project manager to be able to look at their social impact in financial terms. 74

The 30th Special CU-af Seminar 2022 September 2, 2022 Results and Discussion Stakeholder engagement and project dissemination in Nan walking street Nan walking street run by municipality which have more than 50 souvenir shops, food stalls along Nan walking street near Phumin temple. Waste management in walking street is operated, collected by local government, and dispose to landfill within Nan province. The dissemination and communication of the project were done before and during the project. The present waste management practice as well as the problem were gathered and discussed among various key stakeholders in Nan waking street. Several actions and activities were proposed to enhance local practice in waste segregation, reducing single use plastics campaign and banned foam (polystyrene) containers in Nan walking street. The waste that has not been properly sorted, utilized, and recycled were sent to dispose in landfills which is still exist. The mayor, staffs and local respondents in Nan were involved in achieving an effective solid waste management system. The perspective in sustainable waste management was raised in the interviews when discussing improvements and barriers in Nan walking street. Figure 3: Interview and project dissemination with Mayor of Nan municipality, shop owner and staff in Nan walking street Waste composition and waste generation in Nan’s walking street Nan’s walking street (Kad Khuang Muang Nan) happens every Friday - Sunday with decoration in northern-style street food and local souvenirs. The walking street is along the street in front of a local tourist centre and near Phumin temple. The walking street have set the 75

The 30th Special CU-af Seminar 2022 September 2, 2022 eating areas for visitors, shops, and a lot of food stalls (Figure 4). From survey results from June 2020 – January 2022, the number of tourists visit Nan Walking Street varied depending on season, events, and holiday, with reach the maximum number of tourists about 3,000 and the lowest is 500 person per day. The average number of tourists per day is about 1,900 people. The average amount of waste generated in Nan walking street is about 240 kgs per day with reached 730 kg per day (from operating time about 5 hrs from 5 -9 pm). The composition of waste characterization in Nan walking street can be seen in Figure 5. Figure 4: Eating place and waste sorting point in Nan walking street In Figure 5 the results of the waste characterization study are presented. The waste composition indicates that the most common type of waste is general waste (51%) followed by food waste (33%) bagasse containers, plastic bottles, wooden skewer, and others for about the same portion of the weight (2-7%). From our survey, general waste contains clean plastic bags, plastic waste with contaminated by food waste or other types of waste. The plastic (recyclable waste) is sellable in case of fully separated and cleaned. Thus, the following activities were intended to implement the practice in waste segregation at source and replace single use plastic by using locally and environmentally friendly products. 76

The 30th Special CU-af Seminar 2022 September 2, 2022 Figure 5: The results of the waste characterization in Nan’s walking street (Year 2020 - 2021) Activities on waste reduction mechanism designs/campaign to enhance local capability on waste management practice in Nan’s walking street The results are presented below with two sub activities introduced to Nan walking street to enhance local capability in sustainable solid waste management. • Bamboo products replace plastic items such as bags, container and cup holder which produced from community enterprise within Nan province Using local environmentally friendly material replaced single use plastic bag such as plastic bag, plastic food container by introducing bamboo basket, bamboo container and bamboo cup holder to owner of the food stall in Nan walking street. The project set up the borrowing-returning point for bamboo basket in the entrance of the walking street. The total number of 200 bamboo basket, 2000 bamboo container and 3000 bamboo cup holder were brought to Nan walking street which these products were supported community enterprise in Nan province. (Figure 6). 77

The 30th Special CU-af Seminar 2022 September 2, 2022 Bamboo basket Bamboo container Bamboo cup holder Figure 6: Bamboo products were introduced to Nan walking street The results showed that plastic waste generation were reduced in Nan Walking Street. As a result, the total amount of plastic waste was decreased at a rate of 468 kilogram per year with account for greenhouse gas emission reduction 1.23 tonCO2e/year. Moreover, the generation of household income was increase of 33,500 baht or 975 US$/per project implementation. • Enhance capacity buildings as waste segregation at source, communication, education and installation separated garbage bins For this activity, it was delegated for youth participation in waste management system. The youths in Strisrinan school in Nan province competed in designing for Nan identity bin poster. Figure 7 below showed the award winning in the design contest and bin poster in waste segregation points in walking street. 78

The 30th Special CU-af Seminar 2022 September 2, 2022 Design contest for poster of bins in Nan walking street 79

The 30th Special CU-af Seminar 2022 September 2, 2022 Figure 7: Education and activities on how to separate waste and waste segregation at source campaign in Nan walking street At a regular basis, segregation point at Nan walking street were separated in 9 types of waste as utilization, recycling, followed proper disposal process of each waste (general waste, food waste, plastic bottles, plastic cups, plastic bags, plastic cutlery, plastic straw, bagasse container, wooden skewer). The installation garbage separation, and education in waste separation were implemented to tourist in Nan walking street. Public awareness to reduce waste and to increase the utilization of organic wastes and recyclable wastes are benefit from the project. The results 80

The 30th Special CU-af Seminar 2022 September 2, 2022 reported the amount of solid waste reduction to landfill, plastic waste reduction to landfill and increase of plastic recycling rate are 11,334 kg/year, 625 kg/year and 1,526 kg/year, respectively. Table 2 are presented in the economic value, social value, and environmental impact of the project from waste segregation at source and recovered materials. The value of the resources is not only account for material can be recycled, but also the amount of greenhouse gas emission reduction from the project. Evaluation of projects achievement in Nan’s walking street The research implemented two activities to enhance local practice in waste management in Nan walking street. In this section, the project achievement and evaluation were proposed in three dimension as social, environmental, and economic aspects. By using Social Return on Investment (SROI) is a tool to measure the economics aspects. The environmental aspect and social dimensions were presented in the view of pollution prevention, waste utilization and greenhouse gas emission reduction within the scope of this research. From the study, the results can be summarized as the project’s achievement for being further monitoring planning in Table 1 81

The 30th Special CU-af Seminar 2022 September 2, 2022 Conclusion and Recommendation The research intended to strengthen local capability toward sustainable municipal solid waste management: Nan walking street as case study. Stakeholder engagement were one of the key actions to communicate and to enhance waste management practice. The activities on waste reduction mechanism schemes were implemented in Nan’s walking street. The results revealed the economic benefit in term of SROI. The project gained SROI ratio of 1.30 and 2.07. It proved that every 1 baht invested has a benefit of 1.3 and 2.07 baht in activities 1 and 2, respectively. The environmental aspect showed significant number of decreasing municipal waste and plastic waste reduction that can be calculated for GHGs emission reduction = 1.23 (Activity 1) and 29.4 (Activity 2) tCO2eq/year. Social aspect represented the benefit distribution to the community accounting for household income, job creation and achievement of the goal to promote sustainable waste management in Nan walking street. It is recommended that further development of the research as drivers of sustainable city can be: • Project continuity will ensure that any activities or planning are undertakings to proceed even a period of disruption such as Covid-19 pandemic. The projects invest in waste sorting equipment and practice for collecting/separating plastic waste, recyclable waste and other types of waste which should be consistently operating and monitoring the performance including development planning and project expansion accordingly with Local Authority Strategic Development Plan. • Local policy is needed to ban and prohibit single use plastic in commercial areas as in this case study (walking street, vendors, public events, shops). Another impactful policy changes are needed to encourage, provide financial support, and promote environmentally friendly product such as compostable cups, plates and cutlery with increase of local participation, stakeholder engagement, provision of information and education about campaigns in plastic waste management. • An expansion of project areas and implementation of activities on other types of waste especially potential project or activities in significantly reduce greenhouse gas emission such as food waste management project References 1. Suttibak, S., Nitivattananon, V., Box, P. O., & Luang, K. (2005). Enhancing Solid Waste Management Capacity of Local Government Authorities: Review of Current Status in Thailand Asian Institute of Technology, In this study , twenty municipalities were selected as representatives of the current solid waste management i. Waste Management, July, 5–7. 2. Gunawan, Y., Wibisono, B. E., Yudistyana, R., & Putri, D. T. (2021). Economics Development Analysis Journal Organic Waste Management Program Evaluation: A SROI and Action Research Article Information. Economics Development Analysis Journal (3). 3. Community waste management action plan “Clean Province” in accordance with the civil state guidelines for the year 2018, Department of Local Administration. Ministry of the Interior [in Thai]. 4. A guideline for administrators of local government organizations. Integrated Waste Management in the Community, October 2009, Waste and Sewage Division Bureau of Waste and Hazardous Materials Management Pollution Control Department Ministry of Natural Resources and Environment [in Thai]. 82

The 30th Special CU-af Seminar 2022 September 2, 2022 5. Nan Solid Waste Action Plan. 2017. Nan Provincial Office of Local Administration with the Bureau of Nan Natural Resources and Environment [in Thai] 6. Solid Waste Management in Bangkok, Department of Environment Bangkok Metropolitan administration 7. National Policies, Initiatives and Best Practices for Solid Waste Management in Thailand. 2017. ASEAN Conference on reducing marine debris in Asian region Phuket, Thailand 8. Puangsiri and Pharino. (2010). Carbon accounting system from integrated municipal waste management in Thailand: case study Sakhonnakhon province, Thesis, Graduate School, Chulalongkorn University. 9. Maalouf and El-Fadel. (2019). Towards improving emissions accounting methods in waste management: A proposed framework, Journal of Cleaner Production, 206, 197-210. 10. Ai, N., and Zheng, J. (2019). Community-based food waste modeling and planning framework for urban regions, Journal of Agriculture, Food Systems, and Community Development. 83

Characterization and product distribution via thermal conversion of abandoned, lost or otherwise discarded fishing gear (ALDFG) waste Yotwadee HAWANGCHU, Ph.D., Prof. Viboon SRICHAROENCHAIKUL, Ph.D., Duangduen ATONG, Ph.D.

The 30th Special CU-af Seminar 2022 September 2, 2022 Characterization and product distribution via thermal conversion of abandoned, lost or otherwise discarded fishing gear (ALDFG) waste Yotwadee HAWANGCHU, Ph.D.1* Prof. Viboon SRICHAROENCHAIKUL, Ph.D.2* Duangduen ATONG, Ph.D.3* Abstract Abandoned, lost or otherwise discarded fishing gear (ALDFG) is a serious growing concern globally due to the numerous critical impacts on environment and economic. The general treatment of this waste has been incineration for energy recovery and disposed in landfill in which the latter is prohibitive from desperate shortage and a severe damage to the environment. For resource recovery and recycling, thermal conversion route has the great potential to generate fuels, chemicals, and synthesis gas. To analyze the effect of pyrolysis temperature on product selectivity, Fishing net wastes (FN) were pyrolyzed at 400-700 ºC using Py-GCMS instrument. FN were collected from 4 sites in the Gulf of Thailand noted as KFN, LFN, MFN, and UFN. Functional group and thermal decomposition of FN were studied using FTIR and TGA. The pyrolytic products were categorized according to its chemical species. Observed functional group of FN indicated signal of amide linkage (-CO-NH-) of KFN, LFN, and MFN that corresponding to polyamide (PA) as raw materials. While UFN, signal of -CH- stretching of aliphatic structure of polyethylene (PE) were observed. 1Aquatic Resources Research Institute, Chulalongkorn University, Bangkok, Thailand 2Department of Environmental Engineering, Faculty of Engineering Chulalongkorn University, Bangkok, Thailand 3National Metal and Materials Technology Center, National Science and Technology Development Agency, Pathumthani, Thailand 85