Melanoma Cell Detection by Using K-means Clustering Segmentation and Abnormal Cell Detection Technique

Abstract

Melanoma diagnosis in earlier stage is a challenging task, and it is ideally done by the pathologist to automate the detection of melanoma cells. Detection of melanocytes from histopathological image is difficult because melanocytes are very much similar to the keratinocytes. To detect melanoma cells at first, segmentation is required followed by calculation of perimeter of the cells. If it is greater than the threshold value, then it is identified as melanoma cells. K-means clustering unsupervised learning technique is used to segment the image followed by ellipse fitting algorithm to find out abnormally large cells. The centroid is then calculated and marked as nuclei. Experiment is performed on 30 images and is evaluated based on PPV, sensitivity, and NSR. The proposed work achieved 90% PPV, 85% sensitivity, and 13% SNR value.