Ovarian Cyst Classification and Abnormality Detection Using Convolutional Neural Networks

This project aims to develop a deep learning network-based computer-aided diagnostic (CAD) system for the categorization and anomaly detection of ovarian cysts. There are 1319 ultrasound images of ovarian cysts in our dataset which was gathered by multiple operators on a 4D ultrasound machine called the Voluson-S8 from a top obstetrics and gynecological facility. General Electric (GE) produces the Voluson-S8, a device designed especially for scans pertaining to gynecology. Aradiology specialist with over 20 years of experience in gynecology carefully classified every image in the dataset. Five classes of images were created. MATLAB R 2022b's Deep Learning Toolbox is used to create the two deep learning networks, GoogleNet and VGG-19. These two pre-trained deep learning networks are compared on several important parameters discussed in paper. VGG-19's validation accuracy is 74.9%, while GoogleNet's is 92.4%. Compared to VGG-19, GoogleNet is 17.5% more accurate. In the current study, we were capable of categorizing the ovaries into five distinct abnormalities namely Endometriomas, Hemorrhagic Cyst, Hydrosalpinges, Polycystic, and Normal Ovaries. The gynecologist/radiologist, who currently performs the task routinely and has many years of experience, compared the classification results with their analysis. When compared to gynecologists and radiologists, we have discovered that the deep learning (DL) based CAD system exhibits a high degree of accuracy in the identification and categorization of ovarian cysts.