Handling Uncertainty with Fuzzy Lesion Segmentation Improves the Classification Accuracy of Skin Diseases using Deep Convolutional Networks

Often skin disease classification models suffer from confusion due to similar lesion regions with background skin. It has been observed that disease lesions are sometimes non-distinguishable due to similar structure and texture with skin, which leads to misclassification. Segmentation of lesions may help to improve the accuracy of prediction by extracting the region of interest. However, exclusive clustering-based segmentation methods limited handling uncertainty in the lesion regions. Fuzzy clustering methods are built to handle such uncertain homogeneous regions.In this work, we employ Fuzzy C-Means (FCM) segmentation to extract lesion from the diseased skin images. Segmented images are then fed into Deep Convolutional Neural Network (DCNN) for skin disease classification. The comparative analysis over the traditional segmentation techniques demonstrates that the FCM segmentation enhances the performance of DCNN in classifying skin diseases with improved accuracy.