Ultrasound Image Classification and Follicle Segmentation for the Diagnosis of Polycystic Ovary Syndrome

PCOS is a prevalent hormonal disorder that impacts women in the reproductive age bracket. Timely and accurate diagnosis of PCOS is crucial for the proper treatment. To diagnose the presence of PCOS, ultrasound images of the ovaries are widely used by the physicians. Automated detection of PCOS shall reduce the risk of making errors. This study seeks to propose a machine learning classification technique for PCOS detection and to mark the cysts on the ovary using image segmentation. Convolution Neural Network (CNN) architectures such as Inception V3, VGG16, and ResNet are used for classifying images. The model is trained over 781 ovary ultrasound images to distinguish between PCOS and non-PCOS cases. Among the three models used VGG16 model comes with better accuracy. The results of this study show that this approach is effective in detecting PCOS with high accuracy.