Segmentation of Uterine Fibroids Ultrasound Images in Focused Ultrasound Ablation Surgery Based on YOLACT

Uterine fibroids (UF) are the most frequently occurring non-cancerous growths in women. The treatment of UF involves the utilization of focused ultrasound ablation surgery (FUAS), which effectively targets and treats these fibroids. Throughout the surgical procedure, ultrasound images serve as a means to closely monitor and guide the progress of the surgery. Due to the low resolution and the boundaries of the lesions are less clear in the operation. The study involved an analysis of ultrasound images from 1420 patients who underwent FUAS treatment for uterine fibroids. The aim was to identify the most effective real-time ultrasound image recognition method by comparing several deep learning algorithms. In order to accurately register the images, the segmentation results of magnetic resonance images were utilized to indirectly identify the surrounding tissues of uterine fibroids. The main objective of our study is to accurately outline key tissue information such as the bladder, uterus, ureteral balloon, and uterine contour, among others. in ultrasound images to provide data support for registration and tracking. Through the comparison of several deep learning networks for ultrasound recognition, the optimized YOLACT network was selected as the model for recognizing ultrasound images. The recognition accuracy of skin, bladder, catheter balloon and uterus are 95.010 %, 93.486 %, 88.084 % and 92.468 %, respectively. Due to the imaging characteristics of ultrasound images, we used the method of identifying the boundary of tissues and organs around uterine fibroids, and indirectly identified the bladder, ureteral balloon and skin to determine the spatial position of the uterus, which improved the accuracy and real-time performance of ultrasound image segmentation.