Fuzzy C-Means Applied to MRI Images for an Automatic Lesion Detection Using Image Enhancement and Constrained Clustering

In this work, we present a fast and robust practical tool for detection of lesions with minimal user interaction. Particularly, a fuzzy image enhancement is performed on both T1 weighted magnetic resonance (MR), and Fluid attenuated inversion recovery (FLAIR) images to facilitate a better segmentation. We establish a fuzzy system that performs the intensity transformation through the implication method; after, the scalar output obtained from this system is used to separate healthy from the unhealthy structures using constrained fuzzy clustering. An advantage of this lesion detection pipeline is the simultaneous use of features computed from the intensity properties of the image in a cascading pattern, which makes the computation fast, robust and self-contained. We empirically validate our algorithm with large scale experiments using both clinical and synthetic brain lesion datasets, and an 84%-93% overlap performance of the proposed algorithm was attained with an emphasis on robustness with respect to different and heterogeneous lesion types, and its effectiveness in terms of computation time.