Surgical site infection (SSI) is a common (3-5%) and costly (up to $25,000) complication following surgery. Almost 500,000 SSI are diagnosed annually in the US. SSI are primarily diagnosed via subjective means, with limited objective data, contributing to delayed diagnosis, inappropriate antibiotic use and bacterial resistance, and excess health care cost. Mobile phone-based the

rmal imaging (MTI) can provide a rapid, reliable and cost-effective means to gather objective data for SSI diagnosis, during in-person and telemedicine encounters. However, no previous studies have attempted to determine the normal thermal patterns of healthy or pathologic healing. Previous studies at our institution, a large military hospital in the Pacific Northwest, have demonstrated highly accurate MTI interpretation for the assessment of distal perfusion during tourniquet placement and aortic occlusion balloon inflation. In the present study, we assess known infected versus control sterile wounds using MTI in an animal model to gather comparative thermal data as well as prospective human post-operative MTI data across a broad group of surgical types to define normal and abnormal post-surgical thermal patterns using MTI.

The identification of patterns associated with normal and abnormal healing patterns, as displayed by MTI, is an ideal application of deep learning. Deep learning techniques like Convolutional Neural Networks and Deep Belief Networks have been increasingly used in the classification of images in healthcare, particularly in radiology. A novel aspect of our proposal is to adapt advanced techniques like Deep Q Learning for the prediction task. We propose a study to incorporate the use of deep learning in the identification of MTI patterns associated with normal and abnormal healing patterns. Such a study could uncover critical patterns uninterpretable to the human eye or brain. The applications for this solution are extensive and reach well beyond the obvious implications in military trauma care. The continued expansion of telemedicine services would also benefit from such a reliable, fast, point of care solution.


Author: Matthew Eckert

Coauthor(s): Carly Eckert, Rowan Sheldon, Morgan Barron, Matthew Martin

Status: Work In Progress

Funding Acknowledgment: Study funding through US Army Telemedicine and Advanced Technology Research (TATRC), AMEDD Advanced Medical Technology Initiative (AAMTI) grant.