Developing a Multimodal Prototype Device for Tissue Perfusion Monitoring

Repaired tissue must be closely monitored for at least 48 to 72 hours after vascular anastomosis. However, this monitoring is dependent on highly trained nursing staff who are already over-extended. In this work, we develop a multimodal tissue perfusion monitoring system that utilizes machine learning for unattended alert of post operation complications. The prototype unit utilizes thermal IR, RGB, and a laser speckle images and compares their utility for classification of arterial and venous blockage. A ResNet backbone is used with early and late fusion methods for combining information from the various image modalities. Results on a new tourniquet simulation-based tissue perfusion dataset showed that late fusion using the thermal IR + RGB images performed best at classification and highlight the potential for the prototype to be used in a clinical setting.