Segmentation of medical images for the extraction of brain tumors: A comparative study between the Hidden Markov and Deep Learning approaches

Malignant brain tumors are one of the leading causes of death in adults and children. To identify a brain tumor, an MRI image is acquired and analyzed manually by an expert to find lesions. This procedure takes time and the intra and inter expert variations for the same case vary a lot. To overcome these problems, many automatic and semi-automatic methods have been proposed in recent years to help practitioners make decisions. The advent of Deep Learning methods and their success in many applications such as image classification has helped to promote Deep Learning in the analysis of medical images. In this paper, we will present two methods for the detection of brain tumors in medical images. The first is based on Deep Learning through the U-net architecture that has proven its robustness vis-vis the segmentation of images, especially medical images. The results obtained will be compared by a second method that we have published in another article [1], which uses LBP and k-means techniques. The classes found are improved using the Markov method, by calculating the class correlation. The comparison was made on the same BraTS2019 dataset [2], which will give us an idea of the performance of each.