AI Documentation for the Operating Room (ORCA)
Biomedical Engineering
Luisalberto Acosta, Kaden Kobashigawa, Andrew Opstad, Andres Valdes
Abstract
Medical documentation is vital in the healthcare setting; however, completing this documentation is a time-consuming process that prevents medical professionals from focusing on other tasks and assisting a larger number of patients. While many devices on the market have addressed this concern, they are often expensive to purchase and implement. ORCA has developed a design for an AI automatic speech recognition system that automatically generates medical documentation, allowing medical staff to quickly review and validate it after surgery.
Design specifications were based on clinical needs, including high document accuracy, low cost, and ease of use. ORCA’s planned design is an ambient, LLM-powered dictation system that utilizes advanced language modeling to generate electronic health records for healthcare staff to validate automatically. The system integrates a directional microphone attached to the surgical light in the operating room that sends audio data to the LLM. Validation tests will be performed before displaying the generated document to increase system accuracy and ensure documentation does not display incorrect data, such as false positives or true negatives.
While ORCA will not access real medical records to train the model, open-source simulated medical documentation will be used to demonstrate real-world applicability without compromising privacy. Development of the microphone, along with the training and validation of the generated text, will constitute the majority of ORCA’s design process. Ensuring reliable audio data within the noisy operating room environment will be a challenge, as well as maintaining a high level of accuracy, precision, and recall.
When considering materials and other associated costs related to prototype development, estimated expenses in the academic setting are approximately $150–$250. This includes components such as directional microphones, a microcontroller or audio interface, mounting hardware, protective casing materials, a power supply/cables, and the use of cloud AI APIs if increased performance is required.