A HYBRID NETWORK FOR SEGMENTATION OF OVARY AND FOLLICLE IN ULTRASOUND IMAGES

Abstract

Transvaginal ultrasonography (TVUS) is a common method used to treat infertility. Sonographers can use manual measures to evaluate if the ovary and follicles are viable. In contrast, this procedure is particularly time-timewasting and operator-supported. This research offers a unique combination system, CR-Unet, for segmenting the ovary and follicles simultaneously in TVUS. The CRUnet is a basic U-Net that incorporates a spatial recurrent neural network (RNN). It can efficiently realize multi-scale and long-term spatial environments to overcome the assignment restrictions, which include poor picture condition, less difference, periphery ambiguity, and complicated anatomical structures. In addition, we apply a deep supervision approach to improve the efficacy and efficiency of model training. Self-supervision is also utilized to improve segmentation results on a regular basis. Experiments using 3204 TVUS pictures from 219 patient role showed that the recommended strategy surpassed earlier literature techniques for segmenting the ovary and follicles, with Dice Similarity Coefficients (DSC) of 0.931 and 0.885, correspondingly.