DIP-Based Head-CT Skull-Stripping and Brain Tumor Detection Using CNN

A significant achievement has been obtained in the field of ‘Brain tumor detection from MRI images’. Due to noise issues, CT images are rarely used in this regard. However in remote areas, a CT scan facility is quite accessible rather than an MRI scan. This article proposes a CT brain extraction and tumor detection process. The adaptive lower and upper thresholds for intensity slicing have been computed from the original concept of Otsu’s global thresholding method. Whereas the optimal size of the structural element for mathematical morphology is determined through iteration, a CNN architecture inspired by LeNet CNN architecture has been proposed for tumor detection. A combined dataset of 350 images has been prepared from different web repositories for analysis purposes. The brain extraction gives 97.18% sensitivity, 98.35% specificity, and 92.11% Jaccard similarity index. On the other hand, the tumor detection approach gives 100% training accuracy.