MobileNet Neural Network skin disease detector with Raspberry pi Integrated to Telegram

According to Global statistics cancer, one of the skin diseases, accounts for 8.2 million deaths and 14,1 million new diagnosis per year around the globe. An efficient automated approach that can serve the purpose of early referral for skin disease patients is highly needed. A skin disease module which classifies skin cancer lesions has been constructed in this paper. Skin Lesions classified in this module are 7 including Benign Keratosis and Melanoma, classification utilizing machine learning techniques. The skin disease detector employs MobileNet convolutional neural network on Raspberry pi for the classification of skin lesions utilizing the Keras architecture for training. Telegram chat bot is utilized by the user to take a picture and get a prediction output of the input image which can be taken using gadget camera or upload. MobileNet CNN implements the process of Depthwise Separable Convolution which consumes 8-9 times less Computing resources compared to standard convolution. The model achieves top-3 validation accuracy of 0.096 with 0.89 for top-2 accuracy. This Skin disease detector is promising and has the potential of assisting dermatologists in managing the skin disease diagnosis. Furthermore, it will assist ordinary users to pre-access themselves to get the necessary early referral for proper medical attention to diagnose and manage the disease.