Science & Technology Asia

Association Mapping of Early Mortality Syndrome - Acute Hepatopancreatic Necrosis Disease Tolerance in Litopenaeus vannamei

Anong Nimlamai, Supaporn Khanbo, Warodom Wirojsirasak, Kittipat Ukoskit — Mon, 29 Sep 2025 00:00:00 +0700

Early Mortality Syndrome (EMS)-Acute Hepatopancreatic Necrosis Disease (AHPND) is a severe bacterial disease that significantly impacts Pacific white shrimp (Litope naeus vannamei) farming, leading to substantial declines in shrimp production. To mitigate losses caused by EMS-AHPND, molecular breeding presents a promising approach for sustainable disease prevention. This study focused on the identification of single nucleotide polymorphism (SNP) markers associated with EMS-AHPND phenotypes using Genotyping by-sequencing (GBS). SNP markers were identified in fourth-generation selective breeding lines of L. vannamei. A total of 9,504 filtered SNPs were analyzed for their association with EMS-AHPND phenotypes using the Fixed and Random Model Circulating Probability Unification (FarmCPU), accounting for population stratification and cryptic relatedness. Seven SNPs were identified as significantly associated with EMS-AHPND phenotypes, with P-values passing the Bonferroni-adjusted threshold. This study provides a valuable genetic tool for the genetic improvement of EMS-AHPND tolerance in L. vannamei.

Enhancing Fingerprint Recognition System by the Fused Edge Map

Lwin Min Paing, Charnchai Pluempitiwiriyawej, Lunchakorn Wuttisittikulkij — Mon, 29 Sep 2025 00:00:00 +0700

Biometric identification technologies such as fingerprint, facial recognition, and iris or retina scans are widely integrated into modern identity verification systems, including smartphones, computers, and smart home access control. Among these, fingerprint recognition is one of the most extensively used methods due to the uniqueness of ridge patterns in individual fingerprints. In this paper, we propose a fingerprint matching system based on edge detection techniques. Specifically, we utilize three traditional edge detection operators—Canny, Prewitt, and Sobel—to extract edge features from fingerprint images. The proposed system involves four primary steps: image pre-processing, edge detection using the three operators, fusion of the resulting edge maps, and morphological processing to enhance edge features, followed by a decision-making process based on a matching threshold. We introduce a fusion strategy, Fused Edge Map (FEM), that combines the strengths of each operator to generate a more accurate edge representation. To evaluate FEM, we apply two fusion methods: Majority-based Fusion (MF) and Union-based Fusion (UF). Experimental results show that MF achieves a fingerprint matching accuracy of 92.82%, while UF outperforms all individual edge detectors and the MF method, achieving a matching accuracy of 96.25%.

Catalytic System Improvement Through Computational Approaches

Phanchita Wuttikorn, Pinapa Suttiphong, Surachet Thongboon, Kulpavee Jitapunkul — Mon, 29 Sep 2025 00:00:00 +0700

This research aims to develop a predictive model for hydrogen production using a titanium dioxide-based (TiO2) photocatalyst through an artificial neural network (ANN). The primary objective is to enhance photocatalyst design for efficient hydrogen production, supporting the transition from fossil fuels to clean hydrogen energy while reducing carbon dioxide emissions. The project consists of two parts: machine learning and experimental studies, with a primary focus on the first. In Part 1, machine learning is used to build a predictive model for hydrogen production, utilizing TiO2-based photocatalysts. Pearson correlation is applied to select direct and indirect parameters that significantly impact hydrogen production. Data normalization is performed to minimize variations, and the dataset (122 direct parameter samples and 169 indirect parameter samples) is split by using k-fold cross validation and after that into training (80%), testing (10%), and validation (10%) sets before model training begins. The accuracy of the model is evaluated using R squared and Root Mean Square Error (RMSE). Part 2 involves experimental work focusing on methane-to-ethane conversion using the same TiO2-based photocatalyst. The study compares different silver concentrations to determine the most efficient composition for ethane production. While both parts are crucial, the primary emphasis remains on hydrogen production and predictive modeling, as machine learning plays a key role in optimizing photocatalyst design and improving hydrogen yield predictions.

Microaneurysm Localization in En Face Optical Coherence Tomography Angiography Images

Mon, 29 Sep 2025 00:00:00 +0700

A microaneurysm (MA) is a small, round outpouching of a capillary wall in the retina, typically caused by the weakening of the vessel due to diabetic retinopathy. Microaneurysms are generally difficult to detect because they are very small, low-contrast lesions that can be easily obscured by surrounding retinal structures or image noise. This study employed a machine learning method to localize clusters of MAs in en face Optical Coherence Tomography Angiography (OCTA) images. Twelve features were extracted from MA candidates identified through a rule-based method. A support vector machine was then used to filter out non-MA candidates. The density-based spatial clustering of applications with noise (DBSCAN) method was subsequently applied to localize the MA areas. A predicted location is considered correct if it lies within the ground truth MA area. We tested the method on 150 enface OCTA images known to contain MAs and compared it against the rule-based method. The proposed approach significantly improved the average recall of the rule-based method from 48.69% to 59.32%.

Implementing Pressure Sensing Technology for Healthcare Applications

Mon, 29 Sep 2025 00:00:00 +0700

The demand for fitness solutions that are accessible, reasonably priced, and privacyfocused is rising, especially for applications that call for remote monitoring. Yoga is popular, but because of its complexity and range of poses that call for exact posture, it can be difficult to practice remotely. In order to categorize and estimate yoga poses without the need for camera-based monitoring, this project presents a novel pressure-sensing matrix mat (Asana mat) made of Velostat material. The real-time pressure distribution patterns recorded by the Asana mat are analyzed using deep learning models, such as convolutional neural network (CNN) and random forest for pose classification, and hybrid convolutional neural network long short-term memory (CNN-LSTM) architecture for pose estimation. The system is trained on diverse datasets collected from different users, poses, and execution styles to increase robustness. The results demonstrate a high classification accuracy of 98.05% for classifying 10 poses, making the system a non-invasive, user-friendly tool for improving yoga practice even for remote applications. Additionally, for pose estimation, a hybrid CNN-LSTM architecture was created, which achieved a root mean square error of 0.062 for a prediction every 10-sequence length (10 frames). As a result, this privacy- preserving system is advantageous in both therapeutic and home settings. Deep learning models and non-intrusive pressure sensors show promise for a variety of applications, such as personalized fitness coaching, quantitative physical rehabilitation, and healthcare monitoring.

Automated Detection and Segmentation of Choroidal Neovascularization in OCT Using Deep Learning

Mon, 29 Sep 2025 00:00:00 +0700

Choroidal neovascularization (CNV) is a hallmark of age-related macular degeneration (AMD), marked by the growth of abnormal blood vessels beneath the retina that can severely impair vision. Optical Coherence Tomography (OCT), a non-invasive imaging technique, is widely used to detect CNV due to its ability to capture detailed cross-sectional images of the retina. However, the variability in lesion appearance and the presence of artifacts make manual interpretation challenging and time-consuming. To address these limitations, this study explored the application of deep learning models for automated CNV image detection and CNV localization in OCT images. We compared several architectures widely used for segmentation tasks: U-Net, Attention U-Net, DeepLabV3+, DeepLabV3++, Mask R-CNN, and Mask R-CNN+, tested on a balanced dataset of 500 images. Among the tested models, DeepLabV3++ achieved the highest performance with a CNV image detection accuracy of 99.4% and a CNV localization F1-score of 0.80. These findings suggest that deep learning can significantly enhance the efficiency and consistency of CNV diagnosis, paving the way for its integration into clinical workflows to support early screening and treatment of AMD.

Cup-lump Rubber Quality Estimation from Historical Weather Data Using Machine Learning Regressors

Mon, 29 Sep 2025 00:00:00 +0700

Rubber is an essential raw material in many industries, with rubber trees requiring specific weather conditions to thrive. Thailand is a significant hub of rubber producers, particularly for cup-lump rubber, which is used in many products, such as automobile tires. The quality of this rubber is measured by its percentage of dry rubber content (DRC), indicating the usable portion after processing. Traditional DRC percentage methods often rely on human judgment, introducing potential bias. This study proposes a machine learning approach to predict the DRC percentage using historical weather data. We expanded upon earlier research that applied statistical models by incorporating a wider range of weather-based features and modern regression algorithms. Weather data, including temperature, precipitation, wind speed, and sunlight duration, was obtained from the Open-Meteo API and averaged over time windows, ranging between 3 to 50 days. These features are derived from 2,034 in-house DRC percentage records provided by Southland Rubber Company Ltd. The final dataset comprises 124 features. Several machine-learning regressors were evaluated using Scikit-learn, including XGBoost, LightGBM, Random Forest, and others. The XGBoost model achieved the highest performance, with an R² score of 0.7450 and a root mean square error of 3.23%. These results indicate that machine learning can effectively predict rubber
quality based on variable weather trends. This research offers manufacturers a more objective tool for production planning, resource allocation, and quality control.

Development and Evaluation of a Thai Automatic Speech Recognition Model Using the Conformer Model

Siwakorn Kaewwichai, Kwanchiva Thangthai, Pattara Tipakorn, Wasit Limprasert — Mon, 29 Sep 2025 00:00:00 +0700

This research project aims to develop and evaluate the performance of an Automatic Speech Recognition (ASR) system for the Thai language by leveraging the Conformer architecture. Conformers integrate the strengths of Convolutional Neural Networks (CNNs), which effectively capture local acoustic features, and Transformers, which model long-range contextual dependencies. This combination enhances the overall capability of Thai speech transcription. The experiments were conducted using a diverse Thai speech dataset encompassing various accents, speaker demographics, and acoustic conditions. The dataset includes samples from Common Voice, regional dialects, elderly speakers, and audio with background noise from sources such as YouTube and podcasts. Performance evaluation metrics included Word Error Rate (WER), Insertion Error Rate (IER), and Deletion Error Rate (DER), along with model-related factors such as the number of parameters and processing efficiency measured by the Inverse Real-Time Factor (RTFx). In conclusion, the study demonstrates the moderate potential of the Conformer architecture for Thai ASR tasks, highlighting the need for further development. This includes expanding the quantity and diversity of training data to reflect real-world conditions and enhancing model robustness to complex acoustic environments. Moreover, the Fast Conformer model (115M parameters) contains approximately 13 times fewer parameters than comparable Whisper Large models (1.54B parameters) and achieves an Inverse Real-Time Factor (RTFx) of approximately 6400, which is about 44 times faster than a baseline Whisper Large v3 model (RTFx 146). This suggests its strong suitability for streaming and real-time ASR applications.

Chiral Discrimination of MDPV Enantiomers by Modified 𝛽-Cyclodextrins: Molecular Docking and Semi-Empirical Study

Luckhana Lawtrakul, Suttipong Sutsree, Jakkaphat Kaewmun — Mon, 29 Sep 2025 00:00:00 +0700

In this study, molecular docking and PM7 semiempirical calculations were employed to investigate the binding interactions and enantiorecognition of MDPV enantiomers with various methylated 𝛽-cyclodextrins (BCDs), including heptakis(2-O-methyl)-𝛽 cyclodextrin (2-MEB), heptakis(3-O-methyl)-𝛽-cyclodextrin (3-MEB), heptakis(6-O-methyl)-𝛽-cyclodextrin (6-MEB), and heptakis(2,6-di-O-methyl)-𝛽-cyclodextrin (2,6-DIMEB). The docking simulations revealed three distinct orientations of MDPV within the cyclodextrin cavities, with the methylenedioxy and pyrrolidine rings of MDPV adopting different positions relative to the cyclodextrin rims. The calculated binding free energies (ΔG) indicated that while different orientations slightly affect binding affinity, they do not dramatically influence the stability of the host-guest complexes. Further PM7 calculations confirmed stable 1:1 inclusion complexes, with relative heats of formation (Δ𝑟𝐻) ranging from -41.21 to -92.29 kcal/mol. The methylation of cyclodextrins at specific hydroxyl positions played a crucial role in enhancing enantiorecognition. Notably, 6-MEB, methylated at the narrower primary hydroxyl position, exhibited the most effective enantioseparation, while 2,6-DIMEB, methylated at both rims, showed poor enantiorecognition ability. These findings emphasize the significance of selective methylation in modulating the chiral recognition capabilities of BCD derivatives.

A Robust Graph-based Method with Junction Detection and Angle Pruning for Longest Continuous Non-branching Vessel Segmentation in OCTA Images

Napat Tatiyakaroonwong, Haseeb Ali, Pakinee Aimmanee — Mon, 29 Sep 2025 00:00:00 +0700

Optical coherence tomography angiography (OCTA) is a powerful imaging technique for non- invasively visualizing retinal blood flow at high resolution. Accurate vessel segmentation from OCTA images is essential for diagnosing and monitoring retinal diseases. However, segmentation remains challenging due to the complexity of the vessel network, including sharp turns, varying vessel widths, and frequent junctions. Additionally, denoising OCTA images without losing fine vessel structures further complicates the process. This study proposes an enhanced graph traversal method for OCTA vessel segmentation. Incorporating angular-threshold-based pruning and improved junction handling to address common challenges with a comprehensive preprocessing pipeline to denoise the OCTA image while preserving vessel integrity. It also gives a comparative analysis against a baseline graph traversal technique, which extracts all vessel paths without pruning or junction refinement. This research aims to enhance the accuracy of extraction of longest biologically realistic continuous vessel segments from an OCTA image. To evaluate our method, we use a dataset of five OCTA images each comprising of approximately 175 vessel segments and 60 longest vessel strains. Evaluation metrics include false positive rates and qualitative visual comparisons. Visual analysis demonstrates that our pruning technique significantly improves segmentation quality, producing smoother, continuous and biologically valid vessels while reducing spurious branches. Our results yield an F1 score of 0.8488, showing a marked improvement over the baseline model.

Theoretical Investigation of Cyclodextrin Encapsulation to Enhance the Solubility of Ethionamide and Its Synergistic Boosters

Luckhana Lawtrakul, Nilrat Yodngern, Pisanu Toochinda — Mon, 29 Sep 2025 00:00:00 +0700

Multidrug-resistant tuberculosis (MDR-TB) remains a global health challenge, necessitating novel strategies to improve the effectiveness of existing second-line treatments. Ethionamide, a key antitubercular agent, suffers from poor solubility, low oral bioavailability, and formulation challenges due to its crystalline nature. Synergistic booster compounds such as BDM43266 have been developed to enhance ethionamide bioactivation, but they face similar pharmacokinetic limitations. This study investigates the potential of native cyclodextrins (𝛼-CyD, and 𝛽-CyD) to form inclusion complexes with ethionamide and its synergistic boosters using molecular modelling approaches. Molecular docking and complexation energy calculations were conducted to assess binding stability and host–guest interactions. The results reveal that 𝛽-CyD forms the most stable complexes with ethionamide and selected boosters, particularly BDM41907 and BDM41906, due to optimal steric fit and favourable interaction energies. These findings support the use of cyclodextrin-based drug delivery systems to improve the solubility and therapeutic performance of ethionamide and its synergistic boosters in MDR-TB treatment.

Molecular Modeling of Host-Guest Complexes: A Comparative Study of 𝛽-Cyclodextrin, Calix[4]arenesulfonate, and Cucurbit[7]uril with Bicyclic Azoalkane Guests

Luckhana Lawtrakul, Sodara Thao — Mon, 29 Sep 2025 00:00:00 +0700

This study examines the inclusion complexation of three bicyclic azoalkane derivatives (DBH, DBO, DBN) with macrocyclic hosts 𝛽 -cyclodextrin (BCD), calix[4]arenesulfonate (C4S), and cucurbit[7]uril (CB7). Using molecular docking, Hartree–Fock (HF/6-31G(d,p)), and PM7/PCM methods, binding energies, orientations, and electronic properties were evaluated in gas-phase and aqueous environments. Docking revealed negative binding energies with azo groups consistently oriented toward host cavities. HF calculations confirmed thermodynamic stability, showing negative complexation energies and large HOMO–LUMO gaps. PM7/PCM analysis highlighted solvation effects, where water enhanced hydrophobic interactions and stabilized guest encapsulation, especially in CB7. Comparative results indicated CB7 consistently forms the most stable complexes, followed by C4S and BCD, due to favorable cavity size, steric compatibility, and electrostatic complementarity. Overall, the findings provide insights into host–guest recognition mechanisms, offering guidance for designing efficient supramolecular systems with potential applications in molecular encapsulation and drug delivery.

Computational Investigation of Bond Strength in 𝛼-D-Glucose under Strong Electric Fields: Implications for Plasma-Induced Starch Cross-Linking

Withoon Chunwachirasiri — Mon, 29 Sep 2025 00:00:00 +0700

Plasma technology offers a promising, environmentally friendly approach for starch modification, including cross-linking, which can significantly alter its functional properties. While macroscopic changes in plasma-treated starch are observable, the underlying molecular mechanisms, particularly the specific sites of cross-linking, remain challenging to elucidate experimentally. Starch is a complex polysaccharide composed primarily of glucose units. Understanding how individual glucose molecules respond to the plasma environment at the atomic level is crucial for revealing these mechanisms. This study employs Density Functional Theory (DFT) calculations to investigate the effect of strong static electric fields on the bond strength of 𝛼-D-glucose, a fundamental building block of starch. By systematically varying the applied electric field, we aim to simulate a key interaction experienced by starch molecules within a plasma environment. We will analyze changes in key molecular descriptors, such as bond lengths, bond orders, and vibrational frequencies, to quantify alterations in bond strength. We hypothesize that specific bonds within the glucose molecule will exhibit significant and consistent weakening under the influence of electric fields applied in certain orientations and strengths. Identifying these vulnerable locations computationally is expected to provide valuable theoretical insights into the preferred sites for bond cleavage or rearrangement—crucial initial steps in plasma-induced cross-linking—and offer molecular-level clues to complement experimental characterizations of cross-linking behavior in plasma-modified starch.

Does Water Clustering Accelerate Acetylacetone Tautomerization Reaction?

Kaewalee Vesyasirindra, Kaito Takahashi — Mon, 29 Sep 2025 00:00:00 +0700

𝛽-diketone tautomerization is a key reaction in synthetic chemistry that is known to be accelerated in polar protic solvents, such as water. However, the mechanism and number of water molecules for optimal catalytic effect remain unclear. To investigate this aspect, we employed a hybrid cluster-continuum approach using water clusters, (𝐻₂𝑂)_𝑛, 𝑛 = 1 ∼ 4, to simulate tautomerization of acetylacetone (AA). We utilized CBS-Q energies, along with the geometries and solvation energies obtained from B3LYP/6-311+G(2d,2p), using the polarizable continuum model. The addition of one water molecule was shown to lower the activation energy (Ea) by 24 kcal/mol by actively participating in the proton transfer. Additional water molecules can further lower the Ea by 5 to 10 kcal/mol; however, the decrease in Ea saturates at n = 3 water molecules. Water molecules also help stabilize AA with a solvation energy of −6 kcal/mol per 𝐻₂𝑂. These findings highlight the critical role of explicit water clusters in promoting the tautomerization reaction for AA.

Is Change of Spin State Critical for 3d Transition Metal Carbon Monoxide Bonding?

Natthakrij Nipanutiyan, Kaito Takahashi — Mon, 29 Sep 2025 00:00:00 +0700

Transition metal carbonyl (TM-CO) interaction is seen in different areas, including catalysts for CO2 reduction and biological processes involving CO gas. Due to the complex occupation of 3d orbitals, the spin state can change when a TM binds with various gas molecules. In this study, we evaluated the simplest of such spin crossover reactions: 3d TM atom + CO in detail using B3LYP/6-31+G(d,p). Previous studies on TM-CO by Fournier evaluated the dissociation limit and adduct energies and compared spin states. In the present study, we extended the study to include the calculation of the association potential energy curve. We focused on finding the crossing point of two spin states as a function of the TM-C bond length. We also evaluated the relationship between the change of spin state and stable binding between the TM atom and the CO molecule. We found that Sc, Ti, Fe, Co, and Ni + CO are candidates to be spin crossover reactions that change spin upon TM-CO bond formation. Furthermore, among the 3d TM atoms, the most strongly binding TM atoms were Ni, Ti, Fe, and Co, which showed spin state change upon bonding.

Can Electric Field Modulate the Formaldehyde Hydration Reactivity?

Wayu Takemura, Kaito Takahashi — Mon, 29 Sep 2025 00:00:00 +0700

Previous studies have shown that formaldehyde hydration CH₂O + H₂O, producing methanediol CH₂(OH)₂, can be catalyzed by the active participation of water molecules at the water droplet surface. In such cases, water molecules form a hydrogen bonding network that effectively promotes the proton relay. Considering that electric fields can exist at water interfaces, we evaluated whether an electric field can affect the water catalyzed CH₂O....(H₂O)₂→ CH₂(OH)₂...H₂O reaction. Using B3LYP/6-311+G(2d,2p), we evaluated the effect of geometry relaxation when an electric field is applied to this reaction. When a negative electric field is applied along the carbonyl carbon and water oxygen atom, the activation energy decreases, and the reaction becomes more exothermic. Furthermore, the hydrogen bonding network of the CH₂(OH)₂...H₂O complex was disrupted when the field was +0.153 volts/𝐴˚. Thus, we theoretically found that the electric field can have a profound effect on disrupting important hydrogen bonding networks and can affect the reactivity.

DNA Aptamer (LepDapt) Against Lipl32 as A Potential Diagnostic Agent for Detection of Leptospira

Tri Cao Vu, Boonchoy Soontornworajit, Yuthana Tantirungrotechai — Mon, 29 Sep 2025 00:00:00 +0700

Leptospirosis is an infectious disease caused by pathogenic Leptospira spp. To enable
early detection, DNA aptamers (LepDapt) were developed to target LipL32, the most abundant outer membrane protein in pathogenic Leptospira. Among the identified candidates,LepDapt-5a exhibited the strongest binding affinity. Molecular dynamics (MD) simulations show that LepDapt-5a forms a stable G-quadruplex (G4) structure. While the G-quadruplex serves as the primary contributor to LepDapt-5a’s interaction with LipL32, the adjacent double helix enhances binding affinity by increasing the interaction surface. Analysis of perresidue binding energy via MM/PBSA highlights the significant roles of T19 and G24 in target interaction. To assess the impact of these residues, we used AlphaFold3 to predict the 3D structure of the DNA aptamers and docked them to LipL32 via HADDOCK2.4 webserver. MD simulations of all complexes were performed using the AMBER ff14SB and OL15 force fields for protein and nucleic acids, respectively. While the mutants preserved G4 formation, the G24 mutation disrupted the double helix structure but strengthened binding to LipL32. In contrast, mutation at T19 preserved the helical structure but weakened binding. The binding free energy (Δ𝐺) of LepDapt-5a, computed via the MM/PBSA method, was estimated to be –12.45 ± 12.99 kcal/mol. This value is consistent with experimental data of dissociation constants (Kd) of 33.97 ± 5.30 nM.

A Preliminary Exploration of Energy Consumption in Blockchain Applications for Thai Healthcare Networks

Supakit Prueksaaroon, Supakrit Nithikethkul — Mon, 29 Sep 2025 00:00:00 +0700

Thailand’s fragmented healthcare information systems limit interoperability, seamless
data exchange, and nationwide analytics. The Thai Health Data Center uses the “43 -files standard” to aggregate hospital data through 13 regional centers before reaching the central database. However, this centralized model struggles with availability and resilience—critical for sensitive, time-critical healthcare data tied to legal and insurance processes. Ensuring data security, accessibility, and protection against loss is vital, especially as expectations for disaster recovery grow amid risks like outages and natural disasters. Blockchain offers a secure and scalable alternative for managing healthcare data. In national systems where trust exists among participants, permissioned blockchain is suitable, providing controlled access,
stronger security, and greater availability. To limit energy use, the proposed design employs Proof of Authority (PoA), a consensus mechanism optimized for trusted environments. Using CloudSim Plus, the study simulates and compares energy consumption between classical and blockchain-based systems. The architecture comprises 23 authorized nodes—including regional health offices, zoning divisions, insurance agencies, and the Ministry of Public Health. Transactions are modeled as cloudlets executed on Dell PowerEdge XR11 servers, with volume estimated via regression from 2018–2023 data at about 75 million monthly transactions. Results show annual energy use of 15,827.62 kWh for classical systems and 29,410.02 kWh
for blockchain, reflecting a 1.86-fold increase due to added resilience and continuity

Chest X-ray Image Captioning Using Vision Transformer and Biomedical Language Models with GRU and Optuna Tuning

Sakol Patcharapanyawat, Chawanat Nakasan, Chantana Chantrapornchai — Mon, 29 Sep 2025 00:00:00 +0700

Chest X-ray (CXR) interpretation is time-intensive and contributes to radiologist
workload and potential diagnostic delays. We propose a multimodal deep learning framework integrating a Vision Transformer (ViT) for global visual feature extraction, a biomedical pre-trained language model (ClinicalBERT) for domain-specific semantic encoding, and a Gated Recurrent Unit (GRU) decoder for sequential report generation. Images from the Indiana University CXR dataset were converted from DICOM to PNG and enhanced with contrast-limited adaptive histogram equalization (CLAHE); reports were cleaned, tokenized, and augmented. Hyperparameters—GRU size, learning rate, and batch size—were optimized using Optuna. On the test set, the ViT + ClinicalBERT + GRU configuration achieved BLEU-4 = 0.278, METEOR = 0.221, ROUGE-L = 0.434, CIDEr = 0.846, and SPICE = 0.530, outperforming CNN–RNN baselines and remaining competitive with transformerbased approaches while being computationally efficient.

Dragonfly Algorithm-Optimized Time-Varying Synergetic Control for Droplet Positioning in EWOD Systems

Mon, 29 Sep 2025 00:00:00 +0700

This paper presents the design and implementation of a Time-Varying Synergetic
Controller (TVSC) for precise droplet position control in an Electrowetting-on-Dielectric
(EWOD) system. The proposed controller integrates the advantages of synergetic control (SC) and time-varying sliding mode control to enhance convergence speed while eliminating chattering in the control input. The TVSC approach utilizes macro variables derived from a time-varying sliding surface to achieve smooth and stable actuation. To optimize the controller parameters, a meta-heuristic dragonfly optimization algorithm (DA) is employed. The stability of the proposed control scheme is analytically validated using the Lyapunov stability theorem. Simulation studies are conducted to evaluate the performance of TVSC in comparison to conventional SC and Sliding Mode Control (SMC) under both translational and periodic droplet motion scenarios. The results demonstrate that TVSC achieves a faster convergence rate than SC while mitigating the chattering effect inherent in SMC. Additionally, under the influence of external disturbances, TVSC maintains superior robustness and precision in droplet positioning. This study highlights the effectiveness of TVSC in EWOD based microfluidic applications

Potential of Using Ethanol as Fuel in the Transportation Sector

Faosie Sittimont, Thanwadee Chinda — Mon, 29 Sep 2025 00:00:00 +0700

Ethanol is considered a renewable energy source which could be blended with fossil
fuels to reduce carbon dioxide emissions. Since the production of ethanol involves various processes, numerous studies have found that one of the processes significantly impacting climate change is the acquisition of raw materials. Therefore, this research will focus on calculating the carbon dioxide emissions that impact global warming from the processes of obtaining fresh and burnt sugarcane, which are the primary raw materials in the production of molasses and subsequently using molasses to produce ethanol. 1 ton of ethanol was used as a functional unit. This research was analyzed following Thailand Greenhouse Gas Management Organization (Public Organization) and the fifth Intergovernmental Panel on Climate Change report. The results illustrate that the acquisition of fresh cane has the greatest impact
on carbon dioxide emissions per functional unit (kgCO2eq/FU), accounting for 31.20% of the total. Moreover, the acquisition of burnt sugarcane has a high value at 22.29%. However, if we compared to the same quantity proportion, the emissions factor from burnt sugarcane are higher than those from fresh sugarcane due to open burning. These values were calculated based on the use of nitrogen-containing fertilizers, which account for 56.73%. This is the highest proportion in the sugarcane production process. In Thailand, the demand for renewable energy continues to rise. Therefore, this research aims to enhance the necessary database for analyzing LCA related to ethanol, as well as to find ways to minimize potential impacts as much as possible.

Modeling the Spatial Durbin Error Model on Open Unemployment Rate Data in Indonesia 2023

Arbain, Memi Nor Hayati, Siti Mamuda — Mon, 29 Sep 2025 00:00:00 +0700

This study applies the Spatial Durbin Error Model (SDEM) to analyze the factors in fluencing the Open Unemployment Rate (OUR) across 34 provinces in Indonesia in 2023. The model incorporates both direct effects and spatial spillover effects of the independent variables. The Maximum Likelihood Estimation (MLE) method was used for parameter estimation. The spatial weight matrix was constructed using a customized contiguity approach. The results show that the Labor Force Participation Rate (LFPR) and Gender Development Index (GDI) have significant direct effects on OUR, while Population Growth Rate (PGR) and GDI also exhibit significant spatial lag effects. The spatial autoregressive coefficient λ is 0.371, indicating significant spatial dependence in the error term. The model’s AIC value 104.750 is lower than the MLR model, confirming better model fit.

Energy-Efficient Paddy Rice Dehumidification using a Thermosyphon System

Mon, 29 Sep 2025 00:00:00 +0700

Efficient post-harvest drying is vital to maintain paddy rice quality, prevent spoilage, and extend storage life. This study presents a thermosyphon-based dehumidification system, tested with hot air and hot water heating at 60, 70, and 80 ◦C. The system includes a cylindrical drying chamber with automated controls for higher efficiency. Performance was evaluated using drying time, energy efficiency, and specific energy consumption (SEC). Results showed the system reduced paddy rice moisture from 26.65% to the target 14% (d.b.). Drying times with hot air were 128, 76, and 50 hours, while hot water required 104, 62, and 42 hours at 60, 70, and 80 ◦C, respectively. Hot water at 80 ◦C achieved the fastest drying, completing the process in 42 hours. Energy performance analysis revealed the lowest SEC of 89.05 kWh/kg𝑤𝑎𝑡𝑒𝑟 for hot water at 80 ◦C, whereas hot air at 60 ◦C recorded the highest SEC of 1,204 kWh/kg𝑤𝑎𝑡𝑒𝑟 . Overall, the thermosyphon system demonstrated strong potential for balancing drying speed and energy use. The study supports thermosyphon-based drying as a scalable, energy-efficient solution for post-harvest rice management, with hot water offering both rapid drying and efficiency

Assessing Artificial Intelligence (AI) Literacy and Readiness in Thailand’s Workforce: Challenges and Opportunities for Digital Transformation

Vachirawit Kaewsawad, Jerzy Duda — Mon, 29 Sep 2025 00:00:00 +0700

As this paper seeks to examine how Thailand’s workforce readiness for AI technologies can be influenced by AI literacy, anxiety, and acceptance factors, the researcher has conducted an online survey of 318 respondents across multiple Thai industries as well as employing the Structural Equation Modeling or SEM to analyze the relationship between these factors. The modified Unified Theory of Acceptance and Use of Technology (UTAUT) framework has also been implemented and revealed that AI literacy can significantly reduce anxiety (𝛽 = -0.892, 𝑝 < 0.001) while the effort expectancy strongly influences the AI acceptance (𝛽 = 0.370, 𝑝 < 0.001). This suggests that there is a psychological barrier that should be addressed alongside technical skills improvement. The study has also found that AI literacy can positively influence performance expectancy (𝛽 = 0.262, 𝑝 < 0.001) and effort expectancy (𝛽 = 0.327, 𝑝 < 0.001) which enhances AI’s perceived usefulness and ease of use. The author’s model validates ten of the fourteen hypotheses, confirming that the facilitating conditions can significantly impact AI’s acceptance while social influences show positive yet non-significant effects. These findings altogether provided actionable insights for policymakers and organizations to develop and adopt the targeted interventions that can enhance both AI competency and a more optimistic environment towards integration of AI technologies in Thailand’s workforce.

Assessment of Ground Water Quality with Fluoride Contamination and Its Risk Quantification

Vinod Kumar Kushwah, Kunwar Raghvendra Singh, Dharmendra Singh, Gaurav Sharma — Mon, 29 Sep 2025 00:00:00 +0700

Freshwater demand has surged due to urbanization, population growth, and domestic use, while both natural and human activities have degraded water resources. Monitoring water quality and availability is therefore vital. Mathura, a well-known religious and tourist hub in Uttar Pradesh (UP), India, was selected to assess groundwater quality with respect to fluoride (F-). Groundwater samples were collected from seventeen tube wells and hand pumps across the district during 2021–22. Fluoride concentrations ranged from 0.7 to 3.3 mg/L. Notably, 12 out of 17 samples exceeded 1.5 mg/L, surpassing the Bureau of Indian Standards (BIS, 2012) permissible limit for drinking water. Health risk assessment indicated a heightened likelihood of non-carcinogenic hazards from oral fluoride intake, particularly at sampling site SS8. Among the exposed groups, children were the most vulnerable, followed by teenagers and adults. These findings highlight a pressing health concern in the region. The study provides valuable insights for policymakers, offering data to support targeted interventions and management strategies aimed at ensuring safe groundwater supply for domestic consumption and drinking in Mathura.