An automated deep learning models for classification of skin disease using Dermoscopy images: a comprehensive study

With the explosion of advanced Information and Recognition model, primarily the Deep Learning (DL) and Transfer Learning (TL) models, all aspects of recent research have been influenced. The Biomedical image analysis has also been considerably subjective by recent technology involvements, carrying about a pattern shift towards 'automation' and 'error free diagnosis' classification methods with markedly improved accurate diagnosis productivity and cost effectiveness. This paper proposes an automated deep learning model to diagnose the skin disease at early stage by using Dermoscopy images. The complete proposed framework is achieved by evaluating the four pre-trained transfer learning CNN models such as DenseNet121, ResNet50, VGG16 and ResNet18 for classification accuracy on skin dataset images. Also, some pre-processing steps are followed to enhance the accuracy; in addition with various simulation parameters like epochs, batch size and optimizers are studied to find the best model. Analysis is performed with two different batch sizes i.e. 16 and 32 and two different optimizers i.e. Adam and SGD optimizers. The best accuracy is obtained on proposed ResNet50 and ResNet18 model on batch size 32 and SGD optimizer. The best value of accuracy on ResNet50 and ResNet18 model is 90% followed by second best accuracy of 89% on DenseNet121 model. ResNet50 and ResNet18 have obtained sensitivity as 97% and 94% for Melanocytic Nevi disease whereas DenseNet121 model has obtained 94% sensitivity in case of Basal Cell Carcinoma disease. This model can be used for early diagnosis of skin disease and can also act as second opinion tool for dermatologists.