Clinical Skin Disease Detection and Classification: Ensembled VGG

Abstract

Our research work has succeeded in integrating ensembles into VGG image classification technique aiming higher accuracy and performance than existing models. The convolutional layers are apt for feature extraction from images. In VGG classifier three fully connected dense layers are used for classification of class from these outputted features. But we have integrated ensemble methods immediately after convolutional layers for purpose of better classification output. Thus, the output (features of image) of convolutional layers is passed as a separate input to both ensemble methods and fully connected layers of VGG for obtaining the class of image. Final class of image is determined by specific strategy after analyzing outputs of ensemble and VGG fully connected layers. All earlier works focused on skin disease classification. Here, we have also experimented with yolo for detection of location and class of diseases. Skin is considered as the most significant part of the body. But this most significant part of the body is easily subjected to various kinds of diseases that spread throughout skin at a faster pace. Early detection and prevention are needed. Our research work aimed at detecting top 10 common skin diseases with higher accuracy. User can upload a pic in a mobile or cloud application and inbuilt AI algorithms will detect the type of skin disease with higher accuracy and thus offering prevention suggestions at an early stage without doctor intervention.