Brain tumor segmentation using river formation dynamics and active contour model in magnetic resonance images

The human brain is quite complex in structure due to which it becomes quite challenging for a radiologist to differentiate tumor from normal tissues, blood clots, and edema. This paper presents a technique to segment the brain tumor from magnetic resonance images using the river formation dynamics (RFD) algorithm and active contour model. The brain tumor segmentation problem is modeled as a combinatorial optimization problem. It searches the tumor boundary using the active contour model which further uses RFD to search the optimized path in a region. RFD is heuristic optimization algorithm that mimics the way the water leads to the formation of rivers through erosion of ground and deposition of sediments. As a result, the best possible boundary with the minimum value of energy function is obtained. The technique has been evaluated quantitatively and qualitatively on the BrainWeb dataset. The results indicate the remarkable improvement over a few metaheuristic techniques, namely ant colony optimization algorithm, bacterial foraging optimization, particle swarm optimization algorithm, genetic algorithm, firefly algorithm, and cuckoo search optimization algorithm in terms of specificity, sensitivity, dice index, Hausdorff distance, Jaccard index, and accuracy. The presented approach gives continuous and smooth contours with an accuracy of 98.1% and is computationally faster in comparison to other metaheuristic techniques.