DEVELOPMENT OF A DEEP LEARNING APPROACH FOR OVARIAN CYSTS DETECTION AND CLASSIFICATION USING ADARESU-NET
Keywords:
Ovarian Cysts, Deep Learning Model, Adaresunet Model, Guided Bilateral Filter, Confusion MatrixAbstract
Ovarian cysts are common gynaecological conditions that require accurate and timely detection to guide clinical decisions and prevent complications. Early and accurate detection of ovarian cysts is critical for timely clinical intervention and effective patient management. Conventional diagnostic techniques, such as ultrasound imaging, often depend heavily on the expertise of clinicians, making automated detection methods highly desirable. The application of an adaptive residual U-Net (AdaResUNet) architecture for the segmentation and detection of ovarian cysts in ultrasound images was employed for this study. AdaResUNet combines the strengths of U-Net for semantic segmentation with residual learning and adaptive feature recalibration to enhance model performance on complex medical imaging tasks. A balanced dataset of 700 cyst images (400 for training and 300 for testing) were sourced from OASIS and OC400 dataset. The model recorded recall of 98.5%, precision of 98.4%, F1 Score of 98.8%, and accuracy of 99.9%, showing strong performance. The network incorporates attention mechanisms and deep residual blocks to improve feature propagation, gradient flow, and localization accuracy. The results suggest that AdaResUNet can serve as a reliable tool to assist clinicians in the early diagnosis and management of ovarian cysts, potentially improving patient outcomes.
