Detection and Segmentation of Skull Fractures via CNN and U-Net Hybrid Model using Computed Tomography Images

Computer tomography (CT) are now widely used for the diagnosis of head various head injuries. A CT scan slice has a lot of data that can’t always be thoroughly analyzed quickly and precisely enough with standard visual inspection. To help doctors, a computer-assisted skull fracture detection and classification expert system is required. The most widely used deep learning models for image classification are convolutional neural networks (CNNs), as their accuracy and output consistently beat those of other models. This research aims to develop a model with a hybrud model of CNN with a custom classifier, DenseNet-161 and U-Net, that shows an increased efficiency and ease, through detection and segmentation of the CT scan images in the CQ500 dataset. The model achieved an accuracy of 98.37%, AUC score of 0.972, Recall score of 0.88 and Precision score of 0.87. These results represent a notable improvement over other models.