KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/123456789/1

Browse

Search Results

Now showing 1 - 10 of 14

Some of the metrics are blocked by your
consent settings
Deep learning-based object detection of restorative dental instruments with potential implications for workflow automation and infection control in dental supply units
(2026-12-01)
Poomrittigul, Suvit
;
Mittong, Sirawit
;
Thanathornwong, Bhornsawan
;
Suebnukarn, Siriwan
This study presents a proof-of-concept deep learning approach for automated detection and classification of restorative dental instruments on standardized trays, aiming to support workflow automation and infection control in dental supply units. A dataset comprising 1,000 images and 14,000 annotated instances of restorative dental instruments across 14 categories was developed. The YOLOv8 model was trained and evaluated on this dataset using standard object detection metrics, including precision, recall, and mean average precision at IoU thresholds 0.5 (mAP@0.5) and 0.5:0.95 (mAP@[0.5:0.95]). To assess model advancement, YOLOv8 performance was compared against its predecessors, YOLOv5, YOLOv6, and YOLOv7, under identical experimental settings. A session-level data split was implemented as the primary evaluation to minimize data leakage and provide a realistic estimate of generalization across unseen tray configurations. The YOLOv8 model achieved highest mean average precision mAP@0.5 of 95.9% and mAP@[0.5:0.95] of 80.9%, demonstrating robust detection capability under both standard and stringent evaluation thresholds. Across instrument categories, YOLOv8 demonstrated precision ranging from 90.3% to 100% and recall from 80.6 to 98.5%. The findings demonstrate the feasibility of using YOLOv8 for automated restorative dental instrument detection as an early-stage tool for improving supply unit efficiency. While results indicate high detection accuracy and robustness, further validation in diverse clinical environments is needed. Future deployment should incorporate human-in-the-loop verification, audit trails, and error escalation mechanisms to ensure safe and accountable AI-assisted workflows.
Some of the metrics are blocked by your
consent settings
Performance of deep learning models for the classification and object detection of different oral white lesions using photographic images
(2025-12-01)
Khovidhunkit, Siribang on Piboonniyom
;
Phosri, Kunchidsong
;
Thanathornwong, Bhornsawan
;
Rungraungrayabkul, Dulyapong
;
Poomrittigul, Suvit
Computer vision adjunctive technology for oral lesion diagnoses has been developed to detect and identify Oral Potentially Malignant Disorders (OPMDs) and non-OPMDs. The early detection of OPMDs can reduce the risk of oral cancer development, improving the survival rate of the patients. This study aims to evaluate the computer vision technique in the white oral lesion domain within the scope of photographic images. Deep learning techniques for the classification of Convolution Neural Networks (CNNs) and transformer neural networks, and one-stage models of YOLOv7 and YOLOv8 were utilized to classify and detect five classes of OPMDs and non-OPMDs oral white lesions including oral leukoplakia, oral lichen planus, pseudomembranous candidiasis, oral ulcers covered with pseudomembrane and other white benign oral lesions. From the evaluation results of classification, the IFormerBase model achieves overperformance compared to CNN models with accuracy, precision, and F1 score of more than 80% on the test set. The best model for object detection is YOLOv7 with 84.5% mean Average Precision (mAP) at Intersection over Union (IoU) threshold of 0.3 and 74.5% at IoU of 0.5 on the test set. Object detection results reveal promising automatic oral lesion identification, which can be further developed to enhance the lesion screening system.
Some of the metrics are blocked by your
consent settings
The Study of Image Quality Effect on Model Performance for Bacteria Classification
(2025-01-31)
Treebupachatsakul, Treesukon
;
Chomkwah, Wanwalee
;
Tanpatanan, Tananan
;
Poomrittigul, Suvit
One of the key requirements for supervised learning in deep learning model construction is the dataset for training and validation. For gathering the dataset, obtaining various image qualities from different resources is unavoidable, and this has been considered to affect the supervised model performance. This research proposes to demonstrate the effect of image quality involving high and standard datasets obtained from 2 different resources on the performance of models. The various cell characteristics with gram-positive and gram-negative bacteria datasets were challenged for trial. These different datasets were matched and contributed to 5 cases; case 1: train and test with high-quality images, case 2: train with high-quality images and test with standard quality images, case 3: train and test with images of standard quality, case 4: train with standard-quality images and test with high-quality images, and case 5: train and test with combining these two image qualities. Pre-trained CNN models were implemented to prove the purpose with and without stratified K-fold cross-validation. The results of retrained models showed that the high-performance models require high-quality datasets obtained from the same resource as the testing set, which yield more than 90% of all performance evaluation metrics when tested on challenging unseen datasets. This study provides valuable insights for building high-performance models that can be applied to automate microbiology diagnostics, impacting public health and clinical practice.
Some of the metrics are blocked by your
consent settings
Bilingual Audio Depression Identification Model by Machine Learning
(2025-01-01)
Poomrittigul, Suvit
;
Kiatrungrit, Komsan
;
Homsiang, Phanomkorn
;
Treebupachatsakul, Treesukon
The number of depression patients worldwide, particularly in Thailand, is increasing on an upward trend. Depression screening commonly relies on self-report questionnaires. However, these instruments provide subjective assessments. Recent advancements in machine learning technology offer potential improvements in diagnostic accuracy through more objective measures. This study aims to evaluate the effectiveness of machine learning models in classifying depression using a bilingual audio dataset comprising Thai and English languages. Such models have the potential to assist clinicians by providing objective preliminary screening for depression based on vocal analysis, enhancing diagnostic precision and clinical decision-making. Various machine learning models were implemented including KNN, MLP, Random Forest, Decision Tree, SGD, Logistic Regression, SVM, AdaBoost, and Gaussian Naïve Bayes using MFCC-converted audio datasets. The results indicate that machine learning models effectively classify and identify depression even in bilingual audio datasets compared to individual language models, with the highest accuracy reaching 0.95 from MLP and KNN when testing the trained model by a single Thai audio.
Some of the metrics are blocked by your
consent settings
Deep Learning to Classify Bacterial Species in the same Genus
(2024-01-01)
Sheela, Sherin
;
Piang, May Phu
;
Sakorntanant, Sakda
;
Poomrittigul, Suvit
;
Treebupachatsakul, Treesukon
Bacterial strains in the same genus share highly similar morphology, gram-staining characteristics, colony sizes, and spatial arrangements. Therefore, identifying them by deep learning can be quite challenging. This study aimed to assess the classification of 7 species of bacteria from 2 genera of Bacillus and Vibrio by using 8 Convolutional Neural Network (CNN) models. We implemented Python programming along with Keras API within the Jupyter Notebook. The models were constructed and evaluated under unbalanced and balanced datasets by augmentation (rotation, flip, etc.). Transfer learning with fine-tuning, and pre-processing of mixup and label smoothing were also applied to reduce overfitting and enhance generalization. Based on the experimental results on private dataset, the results of InceptionResNetV2 emerged as the top-performing model with a notable accuracy of 82.8%, 88.6% precision, 78.4% recall, and 78.0% F1-score when label smoothing was applied at 0.5 on balanced dataset.
Some of the metrics are blocked by your
consent settings
Proximal Policy Optimization for Crowd Evacuation in Complex Environments - A Metaverse Approach at Krung Thep Aphiwat Central Terminal, Thailand
(2024-01-01)
Chaudhary, Sushank
;
Sinpan, Nitinun
;
Sasithong, Pruk
;
Khichar, Sunita
;
La-Aiddee, Panithan
Efficient crowd evacuation from railway platforms is critical for passenger safety during emergencies. This study introduces a novel dynamic emergency evacuation route generator using the Proximal Policy Optimization (PPO) algorithm within a custom-built 3D simulation environment developed in Unity. We independently created a detailed digital twin of Krung Thep Aphiwat Central Terminal, Thailand's largest train station, and implemented all elements of the simulation, including the Social Force Model, to accurately replicate crowd behaviors and interactions during evacuation scenarios. Through extensive training over 3,000,000 episodes, our PPO-based model achieved significant improvements in evacuation efficiency. The results indicate that in a major emergency scenario, increasing the number of agents in the station reduced the number of remaining passengers from 111 to just 6, highlighting the model's effectiveness. Similarly, in a minor emergency scenario, the average number of remaining passengers dropped from 38 to 1 with the addition of more agents. These findings confirm the model's ability to adapt to different emergency conditions, offering a practical and scalable solution for enhancing evacuation strategies in high-density environments. Furthermore, increasing the agents' sight range also improved evacuation efficiency, with a 20-meter sight range yielding the best results.
Some of the metrics are blocked by your
consent settings
Simulation-Based Headway Optimization for the Bangkok Airport Railway System under Uncertainty
(2023-08-01)
Sasithong, Pruk
;
Parnianifard, Amir
;
Sinpan, Nitinun
;
Poomrittigul, Suvit
;
Saadi, Muhammad
The ever-increasing demand for intercity travel, as well as competition among all modes of transportation, is an unavoidable reality that today’s urban rail transit system must deal with. To meet this problem, urban railway companies must try to make better use of their existing plans and resources. Analytical approaches or simulation modeling can be used to develop or change a rail schedule to reflect the appropriate passenger demand. However, in the case of complex railway networks with several interlocking zones, analytical methods frequently have drawbacks. The goal of this article is to create a new simulation-based optimization model for the Bangkok railway system that takes into account the real assumptions and requirements in the railway system, such as uncertainty. The common particle swarm optimization (PSO) technique is combined with the developed simulation model to optimize the headways for each period in each day. Two different objective functions are incorporated into the models to consider both customer satisfaction by reducing the average waiting time and railway management satisfaction by reducing needed energy usage (e.g., reducing operating trains). The results obtained using a real dataset from the Bangkok railway system demonstrate that the simulation-based optimization approach for robust train service timetable scheduling, which incorporates both passenger waiting times and the number of operating trains as equally important objectives, successfully achieved an average waiting time of 11.02 min (with a standard deviation of 1.65 min) across all time intervals.
Some of the metrics are blocked by your
consent settings
eHealth Internet of Things on Low Power Wide Area Network of Hospital
(2023-01-01)
Manatarinat, Wiraphon
;
Tantatsanawong, Panjai
;
Poomrittigul, Suvit
;
Udomsiri, Sakol
Nowadays, the number of older adults is increasing due to medical advancements and attention to the health of the elderly. Effective healthcare and safety are essential. Therefore, the researcher has an idea to develop a system that can support the care of patients and the elderly safely by detecting falls. However, there are limitations in the contact with the public health system and other technical issues such as high power consumption, cost, and reliability. In this article, we would like to introduce a fictional wearable fall detection system. Algorithms based on threshold are specifically integrated with the Internet of Things (IoT) and a Low-Power Wide Area Network (LPWAN), with nodes monitoring and parsing the data. The server part consists of an application server and a UDP server. The application server is responsible for accessing the data. At the same time, the UDP server is responsible for receiving data and monitoring abnormal data packets sent by the Narrowband Internet of Things (NB-IoT) to develop an efficient and secure data transmission mechanism. For healthcare professionals, eHealth is designed from the server side, which comprises a server application and a client-server architecture. The eHealth system is a digital technology tool and ICT service that connects healthcare providers and citizens to access health services effectively, comprehensively, equitably, and safely for elderly patients and healthcare providers for algorithm validation. We recruit volunteers for events. Daily life and falls the experimental results showed that our presented algorithm could achieve accuracy, indicating the efficiency of our system.
Some of the metrics are blocked by your
consent settings
Preprocessing Technique for Oral Lesion Classification using U-NET Segmentation
(2023-01-01)
Dissorn, Pun
;
Treebupachatsakul, Treesukon
;
Phosri, Kunchidsong
;
Thanathornwong, Bhornsawan
;
Khovidhunkit, Siribang On Piboonniyom
This research aims to use deep learning techniques to segment oral lesions in medical images for use as a preprocessing step in a classification model. Due to the complexity of oral lesions with undefined margins and dynamic shapes, and the limited amount of data certified by dentists, this approach was found to be underfitting and unsatisfactory. To improve the accuracy of the model, a new approach was proposed to segment interferences such as teeth from the images. This allows the model to better focus on the oral lesions. To achieve this goal, we implemented U-net models with different additional Convolutional Neural Networks (CNN) backbones, including DenseNet 121, EfficientNet B3, VGG 19, ResNet 18, SE-ResNet 18, ResNeXt 50, Inception V3, Mobilenet V2 and SE-ResNeXt 50. A segmentation model was trained with five classes of oral lesions: leukoplakia, pseudomembranous candidiasis, lichen planus, ulcer, and other white lesions. The results showed that DenseUNet and EfficientUNet achieved the highest validation and Intersection over Union (IoU) scores of 98% and 92%, respectively. Our proposed approach effectively segmented the interferences from the images, demonstrating the success of these models in handling the approach. Subsequently, a CNN model of DenseNet 121 was employed for classification. The training accuracy achieved 99.1%, while the validation and test accuracies reached 86.1% and 75.5%, respectively.
Some of the metrics are blocked by your
consent settings
The Comparison of Deep Learning Model Efficiency for Classification of Oral White Lesions
(2022-01-01)
Phosri, Kunchidsong
;
Treebupachatsakul, Treesukon
;
Chomkwah, Wanwalee
;
Tanpatanan, Tananan
;
Thanathornwong, Bhornsawan
Oral cancer is one of the top health problems globally. Some white lesions of the oral cavity can develop into oral cancer if not screened and treated immediately. Modern screening technologies are popular for applying deep learning knowledge to screen and classify images. In this study, we used deep convolution neural network (CNN) to classify oral white lesions, ulcers, and normal anatomy using transfer learning, which can reduce training time. Ten pre-trained model of transfer learning including DenseNet121, DenseNet169, DenseNet201, Xception, ResNet50, InceptionResNetV2, InceptionV3, VGG16, VGG19, and EfficientNetB7 are implemented and evaluated. The evaluation of accuracy, precision, F1score, recall, sensitivity, confusion matrix, and AUC-ROC curve are discussed. The trained models of DenseNet169, DenseNet201, and Xception showed the highest testing accuracy of more than 90% and recall of 0.8833. In addition to the precision, F1score, and specificity, the DenseNet169 outperforms at 0.9034, 0.884, and 0.9417, respectively.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results