Feature fusion context attention gate UNet for detection of polycystic ovary syndrome

Abstract Polycystic Ovary Syndrome (PCOS) is a prevalent endocrine disorder affecting women of reproductive age, characterized by hormonal imbalance, irregular menstrual cycles, and ovarian cysts. Traditional diagnostic approaches, which include clinical evaluations, radiological studies, and surgical interventions, are often time-consuming, costly, and not always reliable. To improve the accuracy and efficiency of PCOS diagnosis, this research introduces the Feature Fusion Context Attention U-Net (FCAU-Net) model, leveraging deep learning (DL) techniques. This study makes two key contributions. First, it enhances dataset preparation through Fuzzy Contrast Enhanced (FCE) imaging. Second, it integrates a Feature Fusion Context (FFC) module into the Attention U-Net model, optimizing the extraction of context and position weights from feature maps for better classification performance. An openly available PCOS Ultrasound Image Dataset with 3,800 images was partitioned with 80: 20 to ensure that only original images were used for testing, while augmented samples were exclusively utilized for training to enhance model generalization and robustness. The remaining 3040 images was augmented to form 45,600 images and split into training and validation sets in an 80:20 ratio. The augmented images were processed and tested with several DL models, including DenseNet, AlexNet, VGG19, ResNet, U-Net, and Attention U-Net. Among these, the Attention U-Net initially achieved over 80% accuracy in detecting PCOS. The proposed FCAU-Net, which incorporates the FFC module, demonstrated superior performance, achieving a detection accuracy of 99.89%, significantly outperforming existing DL models. This research highlights the potential of FCAU-Net in providing a more accurate and efficient tool for the diagnosis of PCOS.