Polycystic Ovarian Syndrome (PCOS) Detection Through Deep Learning
Vinayak K. Bairagi, Mousami S. Vanjale, Anushka Dolas, Sanket Zende, Aziz Nanthaamornphong, Siddhant Harpale · International Journal of Engineering Trends and Technology · 2024
Polycystic Ovarian Syndrome (PCOS) is a complicated endocrine condition that is influenced by inherited predisposition as well as environmental circumstances. PCOS is characterized by several symptoms involving cardiovascular disease, hirsutism, acne, hyperandrogenism, and infertility. It is important to diagnose and treat PCOS at the earliest as it can lead to infertility and cardiac issues. Earlier, it was difficult for physicians to distinguish between benign cysts and malignant cysts in PCOS as they typically relied on manual ultrasound tests. This paper introduces a Python based algorithm of Convolutional Neural Network (CNN) to improve diagnostic speed as well as its accuracy. It mainly aims to develop a CNN based ultrasound image classifier that can distinguish between cysts using ultrasound images. This technique simplifies the diagnosis and accelerates the start of the treatment. It can speed up the initial treatment and streamline the diagnostic process, which are crucial in lowering the risk of PCOS related side effects like cardiac problems and infertility. Diagnosis of PCOS can be extremely complicated due to the intricate visuals, which are characterized by follicular overlap and operator variability. This paper discusses a novel approach to overcome these difficulties. The proposed method focuses on CNN-based image processing for feature extraction. The algorithm is trained on a dataset obtained from Kaggle that contains a range of PCOS-related situations to get the discriminative properties required for accurate classification. It is possible to evaluate classification performance using defined metrics by retesting on different datasets. This paper highlights not only the implications for diagnosis but also the significance of early detection of the reproductive system shortcomings. Results demonstrate that CNN exhibits the best testing accuracy of 97.76% and a recall value of 99.27%.