Resnet-Unet-FSOA based cranial nerve segmentation and medial axis extraction using MRI images

ABSTRACT This paper proposes a Resnet-UNet-Fractional Snake Optimization Algorithm (Res-UNet-FSOA) for cranial nerve segmentation. Firstly, MRI images are considered as input, and thereafter preprocessing is conducted utilizing median filtering. In the module of pre-processing, the image enhancement is carried out based upon improved multiscale vesselness that is in identifying local tubular portions of an image. After that, cranial nerve segmentation is done employing Res-UNet, which is an amalgamation of Resnet and UNet. The network is then trained by a devised optimization approach namely, FSOA. The FSOA is proposed by incorporating Fractional Calculus (FC) and Snake Optimizer (SO). Then, start point and end point extraction is executed utilizing deep seeded region growing (DSRG). At last, medial axis extraction is performed using tensor voting and non-maximum suppression (TV-NMS) method. Furthermore, the proposed approach obtained segmentation accuracy of 0.930, Jaccard coefficient of 0.947, and dice coefficient of 0.950.