INTEGRATING WATERSHED SEGMENTATION WITH MACHINE LEARNING FOR BRAIN TUMOR DETECTION IN MRI SCANS

Abstract

Brain tumors are malformed tissues which are present in the brain hindering the normal functions of nerves and can cause serious intellectual degradation or even death when untreated. The timely and accurate diagnosis of these tumors is essential in development of effective interventions, minimizing the mortality rate, and delivery of quality patient services. Of course, the manual segmentation of the tumors in the MRI results is time-consuming, subjective, and likely to cause inconsistency between clinicians. To solve these difficulties, the paper suggests a hybrid segmentation framework associating classical methods of image processing with the methods inspired by machine learning. In particular, gray-level thresholding (Otsu), morphological operations, distance map, and watershed segmentation are combined in order to increase the visibility and outlining of the tumor. Segments layer-by-layer based on the proposed method achieved 92.7 percent segmentation accuracy on the BRATS 2020 dataset, and both Dice, and Intersection over Union scores, matched those of complex deep learning models, without the resource requirements of strong computer systems or large-scale annotations. The method is best to be deployed in resource-limited medical settings because of these features. On balance, the framework offers a comprehensible, high-speed, and effective method to brain tumor detection that has the capability of assisting clinical routine plus subsequent real-time diagnostic implementations.