MURA: Bone Fracture Segmentation Using a U-net Deep Learning in X-ray Images

Developing a robust bone fracture segmentation technique using deep learning is an important step in the medical imaging system. Bone fracture segmentation is the technique to separate out the various fracture and Non-fracture tissues. The fracture can occur in upper extremity parts of the human body like elbow, shoulder, finger, wrist, hand, humerus and forearm etc. X-ray is one of the widely used imaging modality for visualizing and assessing bone anatomy of the upper extremity. X-ray is used in the diagnosis and planning of the treatment for the bone fracture. The problem of computational bone fracture segmentation has gained researchers attention over a decade because of high variation in fracture size, shape, location, variation in intensities and variation textures. Many semi-automatic and fully automatic methods have been proposed and they are becoming more and more mature. A recent technique that is CNN based deep learning gives the promising result of the segmentation. In this Method, MURA (Musculoskeletal Radiographs) database is used. The CNN based U-Net model is trained using the MURA Database. After the training, the Model is tested on the test images. The Evaluation parameters Like Dice Coefficient and Validation Dice coefficient are found out to check the robustness of the technique. The CNN based U-Net architecture gives the training dice coefficient of 95.95% and validation dice coefficient of 90.29% for whole bone fracture segmentation.