Attention-Based Multiscale Deep Neural Network for Diagnosis of Polycystic Ovary Syndrome Using Ovarian Ultrasound Images

Polycystic Ovary Syndrome (PCOS) is a hormonal disorder that impacts a significant proportion of women in their reproductive years. It results in irregular menstrual cycles and elevated levels of androgens, known as male hormones. Women with PCOS often have ovaries that develop numerous small fluid-filled sacs called follicles, but they fail to release eggs regularly. While the precise cause of PCOS remains unknown, early identification and weight loss can help mitigate the risk of long-term complications. In this study, a novel attention-based multiscale convolutional neural network (AMCNN) is proposed for the detection of PCOS. The utilization of dilated convolution aids in preserving the multi-scale features with fewer parameters. The integration of multiscale characteristics is achieved by the attention mechanism, which enhances the importance of features within significant channels. The experimental results demonstrate the superior performance of the AMCNN, surpassing other prominent algorithms with an accuracy of 98.79%, proving its effectiveness for medical industrial applications.