SEGMENTATION OF CORPUS SPONGIOSUM FROM MALE ANTERIOR URETHRA ULTRASOUND IMAGES

In the treatment of male anterior urethral stricture, three-dimensional digital urethral model (TDUM) can he tissue replacement for repairing urethral stricture. The segmentation of corpus spongiosum is key to reconstructing TDUM. In this paper, we extracted Gabor features of ultrasound images and selected different parameters of neural network model for the segmentation of male anterior urethra corpus spongiosum. Two group experiments were conducted: part urethra dataset (PUD) experiment and full urethra dataset (FUD) experiment. In PUD experiment, different parameters of neural network model were selected to obtain the model with best performance. The best performance model was applied in FUD experiment, in which bulbous urethra (BU), penile urethra (PU), bulbar urethral stricture (BUS) and penile urethral stricture (PUS) images were tested respectively. The experimental result shows that our method is robust in segmentation of corpus spongiosum in urethral ultrasound images, and the segmentation accuracy of corpus spongiosum is over 90%. The performance was recognized by clinical ultrasound physicians and urologists. Combined with tissue engineering and 3D printing technology, this method makes the TDUM possible which then provides a new alternative for the treatment of the complex urethral strictures.