Stack auto-encoders approach for malignant melanoma diagnosis in dermoscopy images

Abstract

In this paper we present Computer Aided Diagnosis (CAD) system for malignant melanoma diagnostic based on deep learning using Stack Auto-Encoders. The dermoscopy images are taken from Dermatology Information System (DermIS) and DermQuest, the image enhancement is achieved by various pre-processing approaches. The extracted features are based on Discrete Wavelet Transform (DWT) and texture Analysis. These features become the input to Stack Auto-Encoders (SAEs) for training and testing the lesions as malignant or benign. The experimental results show the rate of accuracy for texture analysis and SAEs is 89.3%, while using DWT and SAEs gives a higher rate of accuracy about 94%. The experimental results prove that the proposed approaches are more accurate than other approaches in this field of melanoma diagnosis.