ABSTRACT
Image segmentation has become a very essential process in medical image processing. It is involved in the application fields of digitized diagnosis for investigating tumors or abnormal tissues in pathology. We have varieties of techniques in image segmentation. One of the unique segmentation technique is k-means; unfortunately it has limitations in segmenting pathological images with scattered populace (group) of tumor nodes like Glomerulosclerosis image of Diabetic Nephropathy, which is a main challenge faced by pathologists for recognizing these regions of abnormalities. Hence we have proposed a new hybrid technique called Genetic K-Means (GKM) Algorithm, which resolves limitations of k-means technique. We have introduced a new hybrid module by superimposing of k-means with evolutionary Genetic Algorithm (GA) for these scattered or distributed nodes of populace (glomerulos) by providing multiple region segmentation with respect to each scattered node. Numerous methods were connected to problems related to the analysis of k-means clustering. Utilization of GKM algorithm is considered in various streams concerning cluster analysis and segmentation performance. In this work, we have explored the GKM utilization to decide on initialization of most excellent clusters and the streamlining of essential parameters like best fitness and mean fitness. Our investigation outcomes the extraordinary capability of GKM enhancement in identifying clusters for segmentation. The segmentation results were encouraging. We have achieved a good success rate of 99% in detecting pathological microscopic glomerulosclerosis image of diabetic nephropathy.