Smartphone-Based Lyme Disease Detection System Combining Real-Time Image Analysis with Geolocation Risk Modeling

This paper presents a novel DenseNet-201-based deep learning system for smartphone-compatible Lyme disease detection, integrating attention mechanisms and privacypreserving federated learning. Our framework achieves $\mathbf{9 4. 0 \%}$ diagnostic accuracy with the assumption of 95 % Confidence Interval and 0.05 significance level on a multi-center dataset of 8,750 patient records from 12 South Asian clinical sites, demonstrating consistent performance across all skin types ($<1.5 \%$ variation) by comparing the existing algorithm Baseline CNN, this achieves $\mathbf{8 8. 7 0 \%}$ of Accuracy Optimized for mobile deployment, the system processes dermoscopic images in 0.8 s on mid-range devices, offering both clinical-grade accuracy and practical utility in resource-limited endemic regions. This approach advances Lyme disease diagnostics by combining robust deep learning with privacy-conscious distributed computation while addressing critical gaps in skin-type inclusivity and real-world applicability than the traditional CNN algorithms.