Complete and Resilient Documentation (CARD) for Operational Medical Environments
[Technical Report, Final Report]
CLEMSON UNIV SC
Pagination or Media Count:
In response to the HITI HFEHRI solicitation for research to demonstrate and validate hands-free electronic health record data entry solutions that will operate reliably in noisy operational environments, alleviate disruption of care for documentation, and prevent loss of documentation, this project studied a system oriented approach to meet these objectives with a platform, CARD, aimed to enable resilient hands-free data collection, preserve complete documentation through stages of care, and present timely information useful for the medical operation. The wearable CARD platform was developed with a GoPro camera, binaural microphones, an NVIDIA TX2 embedded GPU computer, SDXC memory, and Wi-Fi network interface, running Kaldi-based speech recognition with deep neural networks pre-trained on a powerful NVIDIA DGX 8-GPU computer. The entire system can be viewed and controlled by a commander dashboard. The team developed and published several speech enhancement techniques that significantly improved the speech recognition performance, while it continued to develop machine learning algorithms for post recognition error detection and correction.
- Information Science
- Medicine and Medical Research