Use of K-Means Clustering Algorithm for Lesion Segmentation in Dermoscopic Images

The diagnosis of melanoma skin cancer in the early stages is of vital importance owing to the fact that their effects on the prognosis of recovery. The made of these diagnoses are mostly done with visual evaluation of the skin. Therefore, the stated diagnosis of as a result of visual evaluation of the skin is a subjective assessment that because it depends on the doctor’s expertise. In this study, the segmentation of lesion regions in skin images is performed with K-mean clustering algorithm in order to increase the accuracy of diagnosis made by doctors. In the algorithm, the system is tested by selecting K center number 2 and 4 values. A special melanoma data set has been used during the testing. The analysis of the obtained values have been realized using Peak Signal Noise Ratio (PSNR) and Correlation Coefficient (CC) metrics. The performance of this study was evaluated by comparing Canny edge detection and Mean shift algorithm previously implemented by us. In this segmentation process, the selected center number is 4 in the K-average clustering algorithm and in this situation the highest PSNR value is 17,1591dB. According to the metric results, it was observed that the segmentation performed by using the K-mean clustering algorithm which has selected center number equals to the 4 yielded more successful results