Computer-Aided Diagnosis Model for Skin Cancer Detection

This chapter introduces a computer-aided method to detect skin lesion using image features and shape features. Artificial neural networks (ANNs) trained with image features (energy, contrast, homogeneity and correlation) and shape features (asymmetry, border irregularity, color and diameter) in differentiating common nevus, atypical nevus and melanoma using dermoscopy images were described. 120 dermoscopy skin lesion images were collected from online PH2 database. The model was built on a single 3 layers, feed forward back propagation ANNs trained and tested with round robin method. The ANN’s performance was evaluated with receiver operating characteristic (ROC) analysis and chi-square test. The performance was evaluated by comparing total dermoscopy score method with ANNs method. Our result noted that the area under curve (Az) of ROC were 0.807 for differentiating atypical nevus from common nevus, 0.998 for differentiating melanoma from common nevus and 0.959 for i differentiating melanoma from atypical nevus, respectively. This indicated that the ANNs method provided an accurate differential diagnosis in common skin lesions for dermoscopy images.