Transformation in Mobile Health (mHealth) and Telemedicine with Deep Learning Approaches and Data Security
Code: G-1449
Authors: Mahdie Jafari *, Kosar Baroonian ℗
Schedule: Not Scheduled!
Tag: Clinical Decision Support System
Download: Download Poster
Abstract:
Abstract
Background and aims: Mobile Health (mHealth) and telemedicine are pivotal to digital health but face challenges in managing real-time data and ensuring privacy. This study introduces an AI framework integrating wearable devices (e.g., Apple Watch), medical imaging, and patient speech data to enhance remote diagnosis and monitoring. Innovations include interpretable deep learning and Homomorphic Encryption for secure, privacy-preserving analytics. Method: Data from Apple Watch (physiological signals), MedPix (medical images), and the COVID-19 Speech Dataset were analyzed. LSTM models processed wearable time-series data, while Vision Transformer (ViT) detected anomalies in CT scans. Federated Learning and Homomorphic Encryption ensured privacy-compliant training. Results: The proposed framework demonstrated significant advancements in mobile health and telemedicine. Long Short-Term Memory (LSTM) models achieved 93.7% accuracy in detecting cardiac events such as arrhythmia from wearable device data, enabling real-time alerts for critical conditions like atrial fibrillation. For visual diagnosis, Vision Transformers (ViT) outperformed traditional radiomics, identifying lung abnormalities in CT scans with 95.4% sensitivity, particularly in detecting COVID-19 patterns and tumors. Security measures, including homomorphic encryption, ensured full data privacy compliance with GDPR/HIPAA standards, though processing time increased by 20%. To enhance clinical trust, explainable AI (XAI) tools like LIME elucidated model decisions, highlighting heart rate variability and pulmonary nodule patterns as pivotal factors. However, challenges remain. Integrating heterogeneous data streams—such as speech and imaging—requires robust architectures combining Transformers for natural language processing and CNNs for imaging analysis. Additionally, the high energy consumption of wearable devices underscores the need for edge computing optimizations to improve efficiency. Finally, the absence of standardized regulatory frameworks for AI in telemedicine poses a critical barrier, necessitating urgent collaboration among policymakers, technologists, and clinicians to ensure ethical and scalable implementation. Conclusion: This framework merges deep learning, data security, and interpretability to advance mHealth and telemedicine. Future work must address infrastructure integration, energy efficiency, and regulatory alignment through interdisciplinary collaboration.
Keywords
Artificial Intelligence, Digital Health, Mobile Health
