Enhancing Nerve Segmentation Accuracy: A Studyof RCNN and U-Net Models

In today’s rapidly developing medical field every prospect related to patient’s health care in regards to the medicine, rehabilitation, surgery are being automized with the help of Artificial Intelligence, Machine Learning etc. But, the ultrasonic images of the patient’s nerve system are still examined by the doctors manually to check for the alignments in the patients scan report as ultra sound are very crucial to detect the internal injuries or detection of diseases by the imaging of tissues and nerves. So, this paper aims to study the ultrasonic images by analyzing the images by deep learning algorithms comprised of U-Net, Global context modeling and Region based Convolution Neural Networks and other algorithms for the image pre-processing and the model building. The proposed model uses the CNN to visualize the images to check for the pain indexes and the alignments. U-Net is used for segmenting semantic tasks of the nerve segmentation to find out the correct alignment and pattern of the nerves. The evaluation metrics used in this paper are accuracy score, precision, recall and F1 score. So, accurate real time nerve segmentation is essential in ultra sound images and it is essential in various medical applications such as diagnosis and neurological condition monitoring. The major disadvantage of the existing system is that they contain a lot of noise so it cannot accurately segment the nerves for medical treatments like nerve injections i.e., to know the exact pin point location to give the injection. The proposed model is about introducing image screenings to create a generalized approach for analyzing in segmentation of ultra sound images.