Image Segmentation Metrics in Skin Lesion: Accuracy, Sensitivity, Specificity, Dice Coefficient, Jaccard Index, and Matthews Correlation Coefficient

Abstract

One of the main problems in skin lesion detection is image segmentation. This method is essential not only for image processing-based but also for machine learning-based skin lesion detection to improve the performance. The aims of this study are to compare three image segmentation methods (Otsu's thresholding; Coye; and Grabcut) and introduce the Matthews Correlation Coefficient as a metric in assessing skin lesion image segmentation performance. The segmentation performance is not only assess using common metrics, such as accuracy; Dice Coefficient; sensitivity; specificity; precision; Jaccard Index; and Matthews Correlation Coefficient but also object detection to count the number of objects in the segmented image. Using these parameters, the best image segmentation performance is achieved by the Grabcut algorithm. This study confirms the usefulness and informativeness of the Matthews Correlation Coefficient score in assessing the performance of the image segmentation, especially in skin lesion segmentation.