Dermatological Diseases Classification using Image Processing and Deep Neural Network

Abstract

Dermatological diseases rate has been increasing for past few decades. Most of these diseases tend to pass on from one person to another and are also based on visual perspectives, the dermatological diseases of one kind found on one part of the body might look different on another part of the body and diseases of different kinds on one part might look similar on other body parts.Therefore, it should be taken into account at initial stages to prevent it from spreading. So, in this paper, we proposed a system to classify such diseases of 10 different classes containing 5500 images obtained from the Dermnet dataset. The proposed system consists of 2 parts- image processing and transfer learning for training of dermatological images. The image processing part deals with image augmentation and removal of unwanted elements, which is found to be necessary before further processing, else it will affect the output efficiency. And transfer learning part deals with features extractions and fine tuning of pre-trained VGG16 model. The validation accuracy is found of be 74.1% and by further fine tuning is found to be 76.3%, when tested on those dataset. The accuracy can be improved further if more training images data are used.