Abstract:
Skin diseases represent a significant health concern at a worldwide level, especially in areas where accessibility to dermatologists is restricted, the cost of consulting with them is high and there is a delay in seeking medical attention, which in most instances causes late diagnoses and avoidable complications. Due to the swift growth of smartphones and artificial intelligence (AI), increased interest in high-speed, reliable, and privacy-oriented tools exists that can be used to initiate early awareness of skin-health. The proposed project presents the SkinCare: An AI Guide to Skin Wellness, a mobile app that will analyze skin-condition images in real-time using a quantized TensorFlow Lite model and provides intelligent guidance to users by training a lightweight conversational assistant named Gemini Flash-Lite Pro that will assist users in taking high-quality images, interpreting AI outputs, and offering them safe educational advice. All processing is done on the device to guarantee high privacy and low latency with an average of 0.8 seconds inference time on mid-range Android machines. The HAM10000, ISIC Archive, and Fitzpatrick17k datasets were used to train the AI model, with the preprocessing techniques of normalization, augmentation, and class balancing to make the AI model more robust in different skin tones and under various lighting conditions. EfficientNet-Lite and Optimized MobileNetV3 architectures recorded a competitive accuracy of 8790 percent with a much smaller model size to deploy efficiently on mobile devices. To simulate the actual reality of usage, the development and testing was done in Visual Studio Code upon physical Android devices. Feedbacks on the system were reported to be very satisfying by pilots in terms of the clarity, responsiveness, and user-friendliness. In general, the app confirms the efficiency of integrating edge detection with chatbots into a fast, convenient and privacy-aware solution that can raise skin-health awareness in the early stages and increase the availability of preliminary dermatological consultation, particularly, in communities with lower resources and limited access to them.
