Multi-Class Skin Diseases Classification Using Deep Convolutional Neural Network and Support Vector Machine

Globally, skin diseases are the fourth leading cause of non-fatal disease burden. Both high and low-income countries suffer from this burden; indicates the prevention of skin diseases should be prioritised. In this research work, an intelligent diagnosis scheme is proposed for multi-class skin lesion classification. The proposed scheme is implemented using a hybrid approach i.e. using deep convolution neural network and error-correcting output codes (ECOC) support vector machine (SVM). The proposed scheme is designed, implemented and tested to classify skin lesion image into one of five categories, i.e. healthy, acne, eczema, benign, or malignant melanoma. Experiments were performed on 9,144 images obtained from different sources. AlexNET, a pre-trained CNN model was used to extract the features. For classification, the ECOC SVM classifier was used. Using ECOC SVM, the overall accuracy achieved is 86.21%. 10-fold cross validation technique was used to avoid overfitting. The results indicate that features obtained from the convolutional neural network are capable of enhancing the classification performance of multiple skin lesions.