Bangkok: Chiang Mai University (CMU) is using AI to enhance the accuracy of disease diagnosis, developing artificial intelligence to help classify endometriosis lesions from laparoscopic images.
According to Thai News Agency, a research team from Chiang Mai University, led by researchers from the Department of Computer Science, Faculty of Science, in collaboration with the Department of Obstetrics and Gynecology, Faculty of Medicine, has successfully developed a deep learning system for automatically classifying endometriotic lesions from laparoscopic images. This aims to support more accurate, faster, and more efficient diagnosis.
The research designed a structured AI system workflow using the GLENDA dataset, a database of laparoscopic images for endometriosis. The data was systematically improved through image resizing, data standardization, cropping, and enhancement to ensure consistency and suitability for model learning. Furthermore, the research team applied various data augmentation techniques such as image rotation, flipping, distortion, zooming, and displacement to enhance the model's ability to analyze new data and overcome limitations imposed by the limited dataset size. In model development, the dataset was divided into training, validation, and testing sets, and 5-fold cross-validation was used to increase the reliability of the experimental results. The performance of three AI models using Transfer Learning architectures-VGG19, ResNet50, and InceptionV3-was compared and evaluated using benchmarks including Accuracy, Precision, Recall, and F1-score.
The study found that all three models demonstrated high lesion classification performance and consistent results. InceptionV3 performed best, achieving an accuracy of 93%, surpassing ResNet50 (91%) and VGG19 (89%). Furthermore, it achieved the highest precision (93%), recall (94%), and F1-score (93%) in the experiment. These results were validated via 5-fold cross-validation, reflecting the model's robustness and ability to effectively predict new data. Simultaneously, learning curve analysis showed stable learning, low overfitting, and accurate classification, consistent with the Confusion Matrix and AUC-ROC curve analyses, highlighting the model's strong classification potential.
The research confirms that the proposed deep learning system development approach can significantly increase the accuracy of lesion identification compared to previous approaches and has the potential to be developed into a reliable diagnostic support system for healthcare professionals.
Beyond advances in medical technology, this research offers benefits in multiple dimensions. It increases the chances of faster and more accurate detection of endometriosis, supports clinical decision-making, and reduces reliance on specialists, thereby increasing the likelihood of patients receiving appropriate treatment in the early stages. Socially, rapid diagnosis reduces the risk of chronic pain, infertility, and mental health impacts, leading to improved quality of life for female patients. Economically, AI systems can help reduce medical costs, decrease the need for unnecessary surgeries, shorten diagnostic times, and alleviate the workload of medical personnel. Academically, this research reflects the potential of applying Transfer Learning, Data Augmentation, and Cross-validation techniques to limited, specialized medical datasets, and presents a framework for developing systems that can be applied and further developed in future research on medical image analysis using AI.