Classification of Dermoscopy Images Using Naïve Bayesian and Decision Tree Techniques

Malignant melanoma is one of the most dangerous types of skin cancers, which may grow on any part of the body. Medical Informatics utilized computer technology such as Computer Aided Diagnosis (CAD) to diagnose the disease. Many researchers developed CAD systems for melanoma diagnosis. Early diagnosis of melanoma is a main strategy to reduce melanoma-related deaths. This paper presents intelligence techniques namely, Naïve Bayes and Decision Tree to diagnose malignant melanoma. Dermoscopy images are taken from Dermatology Information System (DermIS) and DermQuest, image enhancement is achieved by various pre-processing techniques. The extracted features are based on hybrid Discrete Wavelet Transform (DWT) and Principle Component Analysis (PCA), and texture features. These features become the input to the various classification techniques like Naïve Bayes and Decision Tree to classify the lesions as malignant or benign. The results show that rate of accuracy of using hybrid DWT and PCA with Decision Tree is 92.86%, while Naive Bayes gives a higher rate of accuracy of about 98.8%. The results indicate that the Naive Bayes is better than decision tree, because it has shown excellent diagnostic accuracy; also the results show that the hybrid DWT and PCA features are more effective in improving the accuracy than the texture features for melanoma diagnosis. The comparative results indicated that the proposed techniques have excellent accuracy than the other techniques in this field of melanoma diagnosis.