Deep learning-based ai integration in dual mobile applications for Skin disease detection and doctor-patient appointment management

Abstract

Skin diseases range from minor conditions to serious cancers like melanoma, where early detection is crucial for effective treatment and survival. This thesis presents a comprehensive AI-driven system for skin disease classification using the HAM10000 dataset, which includes over 10,000 dermatoscopic images across seven lesion categories. At its core is a specially designed Convolutional Neural Network (CNN) enhanced with residual blocks and attention mechanisms, trained on a Colab Pro A100 GPU. The model outperformed popular pretrained networks—ResNet50, DenseNet121, and EfficientNetB0—achieving a validation accuracy of 85.08, while the best pretrained model reached only 59.63. For practical deployment, the model was converted to TensorFlow Lite and embedded into two cross-platform Flutter Firebase mobile apps: a Patient App for AI-based skin image analysis and appointment booking, and a Doctor App for appointment management and AI-assisted diagnosis. This work delivers an efficient, scalable solution for early skin disease detection and smart healthcare support