Shadowed Clustering for Speech Data and Medical Image Segmentation

Abstract

The paper presents a novel application of the shadowed clustering algorithm for uncertainty modeling and CT scan image segmentation. The core, shadowed and the exclusion regions, generated via shadowed c-means (SCM), quantize the ambiguity into three zones. This leads to faster convergence and reduced computational complexity. It is observed that SCM generates the best prototypes even in the presence of noise, thereby producing the best approximation of a structure in the unsupervised mode. A comparison with rough-fuzzy clustering algorithm reveals the automatic determination of the threshold and absence of externally tuned parameters in SCM. Experiments suggest that SCM is better suited for extraction of regions under vascular insult in the brain via pixel clustering. The relative efficacy of SCM in brain infarction diagnosis is validated by expert radiologists.