CNN Based Automatic Segmentation of Scaling in 2-D Psoriasis Skin Images

In India, over 3 to 4 % of people are affected by chronic proliferative disease known as Psoriasis. The patchy rashes, small scaling spots, cracked skin, itching, burning or soreness are the common signs and symptoms of psoriasis. These signs and symptoms vary depending on the type of psoriasis. The various types of psoriasis are plague psoriasis, nail psoriasis, Guttate psoriasis, inverse, pustular and erythrodermic psoriasis. These psoriasis affects the skin and, in some cases, the fungal infections will trigger the disease. To evaluate psoriasis severity, various methods are used to monitor the therapeutic response. In this paper, Principal Component Analysis (PCA) and rigid transformations are used for the automatic segmentation of psoriasis. Convolutional Neural Network (CNN) which comprises of convolutional layer, ReLU activation layer, max pooling layer and fully connected feed forward network are used for the classification of skin images. The feature map is extracted from the input images by convolution operation. These feature maps are obtained to train the neural network model to classify the images. The performance metric of the model is calculated after training the model with input images and the model performance varies depending on the type of images. The accuracy, specificity, sensitivity, F1 score are determined to find the best model for evaluation.