• Delirium is a major contributor to morbidity and mortality in the inpatient setting. Also known as acute brain failure, delirium prolongs hospital stays and stresses scarce nursing resources. The altered mental status associated with delirium often can be difficult to distinguish from baseline chronic decline due to dementia. For all these reasons, monitoring technologies for delirium have potential to improve patient outcomes and clinical operations. 

In this blog post, I review LookDeep Health’s IRB-approved collaboration with Oregon Health Sciences University (OHSU) on AI-enabled computer vision monitoring of trauma patients at-risk for delirium. The study was led by psychiatrists who frequently consult for trauma patients to help with diagnosis and management of delirium. Trauma patients are at increased risk for delirium, highlighting the need for better monitoring. Additionally, consulting psychiatrists see patients on every medical and surgical service in the general hospital; always-on computer vision monitoring can give bedside awareness of their patients even when they can’t physically be there. 

AI-enabled computer vision monitoring of the overnight period 

Our computer vision analysis resulted in several data streams, including patient activity and bedside activity by staff and visitors (a potential metric for patient acuity). In addition to numerical data, image-based readouts were generated, such as intelligent video summaries focused on peak activity intervals and heat maps of patient motion. Image-based readouts were generated from video that was de-identified using full-frame blurring. 

One striking aspect of the data was the ability to characterize the overnight period, as demonstrated in the patient story for an 80-year old on the trauma service. She was admitted for non-surgical management of rib, spinal, and hip fractures after a motor vehicle collision. AI-assisted video actigraphic analysis showed significant nighttime patient activity with rest-wake inversion, demonstrated by video summaries generated of the nighttime period as well as by motion “heat maps” that visually display patient activity over a given time interval. AI analysis also revealed significant bedside activity through the daytime and nighttime on most days, consistent with frequent staff visits seen on video summaries. 

The importance of AI in computer vision analysis can be appreciated in the figure below, which is populated with different kinds of readouts generated by analysis of the patient discussed above. Video itself generates massive amounts of data that need to be refined in order to become actionable. AI enables automation of this processing. Further, as AI improves, analyses of increasing utility can be made, assisting clinicians with new knowledge and wisdom to improve patient care and clinical workflows.

Physician consults, virtual medicine, and AI data

As mentioned above, consulting psychiatrists see patients throughout the hospital. As with other consulting physicians, their specialized expertise is essential for optimal and timely diagnosis, management, and treatment of acute illness. Indeed, AI-enabled computer vision, when combined with always-available virtual/tele-medicine, has the potential to significantly decrease the morbidity and mortality associated with delirium. It also has the potential to avoid prolonged hospital stays characteristically associated with delayed diagnosis of delirium. 


OHSU authors
Jonathan R. Floriani, MD
Amela Blekic, MD, DFAPA 

LookDeep Health authors

Michael A. Choma, MD, PhD 
Narinder Singh, MBA, MTM 
Paolo G. Gabriel, PhD 
Laura Urbisci, PhD 
Dewei Hu, MS 
Tyler P. Troy, PhD