Improving the diagnostic accuracy of dysplastic and melanoma lesions using the decision template combination method

Abstract

Background/purpose: Melanoma is the most dangerous type of skin cancer, and early detection of suspicious lesions can decrease the mortality rate of this cancer. In this article, we present a multi-classifier system for improving the diagnostic accuracy of melanoma and dysplastic lesions based on the decision template combination rule.

Methods: First, the lesion is differentiated from the surrounding healthy skin in an image. Next, shape, colour and texture features are extracted from the lesion image. Different subsets of these features are fed to three different classifiers: k-nearest neighbour (k-NN), support vector machine (SVM) and linear discriminant analysis (LDA). The decision template method is used to combine the outputs of these classifiers.

Results: The proposed method has been evaluated on a set of 436 dermatoscopic images of benign, dysplastic and melanoma lesions. The final classifier ensemble delivers a total classification accuracy of 80.46%, with 67.73% of dysplastic lesions correctly classified and 83.53% of melanoma lesions correctly classified.

Conclusion: The results show that the proposed method significantly increases the diagnostic accuracy of dysplastic and melanoma lesions compared with a single classifier. The total classification rate is also improved.