Application of MP-YOLO for Segmentation and Visualization of Ovarian Ultrasound Imaging

Polycystic Ovary Syndrome (PCOS) affects ovaries. If untreated, it can lead to complications such as infertility, cancer, and depression. Currently, the main diagnostic method is medical ultrasound imaging. However, the use of ovarian follicle ultrasound imaging with ultrasound scanning to capture images often results in blurry images and the presence of irregularly sized patches. These factors directly impact the analysis and judgment of medical professionals. To address this issue, we propose an integrated framework combining YOLO and morphology. The framework consists of three stages: YOLO training stage, the detection and extraction stage, and the visualization stage. Firstly, the YOLO algorithm is used to accurately identify the ovarian follicle regions. Then, in combination with morphology, the framework extracts the relevant features of interest from the images. Finally, various visualization techniques are employed to assist medical professionals in the analysis and diagnosis of PCOS. The results of the study show that YOLO achieves an overall performance F1-score of 87.33% in detecting ovarian follicles, representing a 50.83% improvement compared to traditional morphology methods. Moreover, by incorporating morphology, the framework enables automatic selection of the ovarian follicle ROI (Region of Interest) with an average Intersection over Union (IoU) value of 94.63%. This indicates that the automatically selected ROI closely matches the ROI annotated by experts, thereby enhancing visualization techniques such as physical coloring, virtual colorization, contour enhancement, and texture enhancement. These advancements assist in precise medical work and improve the analysis and diagnosis of PCOS.