TLDR: A new portable, edge-enabled Electronic Health Record (EHR) system has been developed to provide essential healthcare services, like non-invasive anemia screening, in remote and resource-limited environments. It operates offline, secures patient data with AES-256 encryption, and uses AI models optimized for low-power devices, demonstrating a viable solution for decentralized healthcare. The system emphasizes low-cost deployment, modularity, and data privacy compliance, addressing critical barriers to digital health adoption in disconnected settings.
Access to essential healthcare services remains a significant challenge in remote and underserved regions globally. Traditional Electronic Health Record (EHR) systems and cloud-dependent mobile health solutions often fall short in areas lacking stable power, reliable internet, and robust infrastructure. These limitations highlight a critical need for self-sufficient, portable healthcare technologies that can deliver medical support directly where it’s needed most, regardless of infrastructure availability.
Researchers have developed an innovative portable, edge-enabled Electronic Health Record platform designed specifically for these resource-limited environments. This system prioritizes autonomy, resilience, and adaptability, operating independently of persistent internet connections. It handles on-device data storage and computation, and is deployable on low-power, cost-effective hardware.
The platform’s core design embraces constraints as drivers, focusing on an offline-first architecture where all data processing, storage, and diagnostic inference occur locally on the device. While cloud synchronization is supported, it remains strictly optional, ensuring reliability in disconnected settings. The system runs on small-form factor embedded devices, such as the NVIDIA Jetson Nano, balancing computational needs with energy efficiency.
As a compelling use case, the platform integrates a non-invasive anemia screening module. This module leverages fingernail pallor analysis, a clinically recognized indicator for anemia. The system employs a YOLOv8n-based model for accurate nail bed detection, which has been optimized through quantization to significantly reduce inference latency while maintaining high accuracy. For hemoglobin level estimation, a Random Forest model was chosen for its robustness and efficiency, trained on a dataset of 250 patient cases.
Data security and privacy are paramount in this design. All patient records are encrypted using AES-256, aligning with stringent standards like HIPAA and GDPR. Access to EHR data is governed by a Role-Based Access Control (RBAC) system, ensuring that only authorized personnel can view or modify patient records. These security measures operate with minimal performance overhead, allowing for efficient data handling on edge devices.
A user-friendly web-based dashboard interface, built with ReactJS and FastAPI, allows healthcare workers to register and manage patient records, visualize anemia screening results in real-time, and track longitudinal health trends. This interface also supports offline operation through local caching mechanisms, ensuring uninterrupted functionality even without network connectivity.
During performance evaluations on NVIDIA Jetson Orin Nano platforms, the anemia screening pipeline achieved an average inference latency of 28 milliseconds per image, with a total end-to-end response time of approximately 42 to 58 milliseconds per patient. Power consumption remained under 7W during active inference cycles, validating the system’s ability to deliver real-time diagnostic support in field conditions.
While promising, the researchers acknowledge certain limitations, including the relatively small dataset used for training the anemia screening model, which may limit its generalizability across diverse populations. The current implementation focuses solely on anemia screening, and broader diagnostic capabilities would require additional data and validation. Future work aims to expand the platform to support additional screening modules, explore deployment on even lower-cost edge platforms, and investigate the integration of renewable energy sources to enhance system autonomy in off-grid environments.
Also Read:
- AI Model for Anemia Detection: Balancing Performance and Efficiency on Portable Devices
- DailyLLM: Generating Rich Activity Logs from Your Smartphone and Smartwatch
This work demonstrates a scalable approach to enhance portable health information systems and support frontline healthcare in underserved regions, contributing to the global mission of improving healthcare accessibility and reducing health disparities worldwide. For more details, you can refer to the full research paper here.
