MobileNet Neural Network Skin Disease Detector with Raspberry PI Integrated to Telegram

Abstract

Skin diseases with their malignant Melanoma and benign Keratosis are two major global health issues that affect people around the world. The timely identification of skin diseases remains essential for qualified medical treatment which delivers better results to patients. The absence of available clinical diagnosis tools results in prolonged medical examinations and elevated death counts among patients. A complete system for skin disease detection presents an integration of MobileNet Convolutional Neural Network (CNN) detection and Raspberry Pi and Telegram realtime monitoring capabilities and notification functions. The MobileNet CNN operates with a lightweight structure that employs Depthwise Separable Convolution to reach processing complexity reduction rates of 8-9 times lower than those of standard convolutional networks. This implementation enables skin lesion classification through transfer learning which re-trains a MobileNet model that already exists for skin lesion specialization. The system provides precise and reliable predictions that achieve validation accuracy of 0.96 for top-3 results and 0.89 for top2 performance when it categorizes skin lesions into Melanoma and Benign Keratosis classes. The Raspberry Pi enables local image processing operations which allows for a budget-friendly system deployment in areas with limited resources. The real-time notification system along with classification outcomes reaches both users and healthcare personnel through Telegram to provide them with quick access to medical assessment. Through this integrated approach detection becomes more efficient at the same time it provides improved accessibility. The proposed detection system provides a strong and budget-friendly technology solution which scales for early skin disease diagnoses worldwide. The future development of the system will focus on expanding the available dataset for better accuracy as well as adding more diagnostic elements that will result in complete skin health evaluation.