Publication

Abstract : Polycystic Ovary Syndrome (PCOS) is a hormonal disorder characterized by elevated levels of androgens produced by the ovaries, leading to the formation of ovarian cysts in women of reproductive age. This condition is often associated with irregular or absent menstrual cycles and the presence of multiple small cysts within the ovaries. Diagnostic methods for PCOS typically include ultrasound imaging, blood tests, and physical examinations to detect follicular stimulation and related symptoms. This study explores the application of deep learning-based image analysis techniques for the identification of PCOS. Ultrasound imaging offers valuable insights into ovarian size, volume, and the number of cysts present. These images often exhibit fluid-filled or blood-filled cystic structures. To analyze and differentiate ovarian features in ultrasound scans, various image processing technique—ssuch as thresholding, region marking, and deep learning models—were employed. Specifically, models like ResNet with transfer learning, ResNet combined with Random Forest, standalone ResNet, VGG with XGBoost, and MobileNet with transfer learning were evaluated. Among these methods, MobileNet with transfer learning, integrated with the Inception V3 model, achieved the highest accuracy of 99.68% in classifying normal and abnormal. The final model was successfully deployed on a web-based interface, offering a potential diagnostic aid for healthcare professionals.