Evaluation of Deep Learning Techniques in the Diagnosis of Polycystic Ovary Syndrome
Elifnur Erdemir, Çağatay Berke Erdaş · 2023
Polycystic ovary syndrome is an endocrine disorder that causes infertility. In addition, polycystic ovary syndrome increases the risk of type 2 diabetes, gestational diabetes, venous thromboembolism, cerebrovascular and cardiovascular disease, and endometrial cancer. The high prevalence of this disease and it’s associated with other health complications place an excessive burden on the health system. In this study, deep learning approaches for detecting polycystic ovary syndrome from ultrasound images were investigated to shorten the diagnosis time and reduce the workload of doctors. Xception, ResNet-152 and DenseNet-201 models were trained and tested without data augmentation, with data augmentation and transfer learning. According to the results, DenseNet-201 model gave the highest accuracy value with 99.48% when data augmentation was made. When the transfer learning technique was applied, it was observed that the Xception model reached 98.44 %, showing an increase in accuracy of 7.79% compared to training and testing the Xception model with raw data. Although the ResNet-152 model showed low performance compared to other models, it was observed that the accuracy of 75.58% without data augmentation increased to 83.12% when data augmentation was applied. Based on these results, it has been revealed that data augmentation and transfer learning techniques will increase the diagnostic performance of polycystic ovarian syndrome.