James Wrobel

Personal Information
Title Associate Professor
Expertise Wound Healing
Institution University of Michigan-Ann Arbor
Data Summary
Grants/SubContracts 1
Progress Reports 1
Publications 1
Protocols 0
Committees 1


Image Processing and Machine Learning for Prediction of Wound Healing
There is a fundamental gap in current wound practice for measuring changes in wounds over time and subsequent clinical decision making. Current standards for assessing wounds involve qualitative description of the wound base and measurement of width and length using a scalpel after debridement. Rarely do clinicians perform the next step of estimating surface area changes over 2 or 4 weeks, important prognostic milestones for wound healing. Modeling wounds as rectangles is also inaccurate. In addition, wound base features provide valuable prognostic information. We have developed automated wound segmentation and machine learning models that are highly accurate in predicting wound surface area, infection, and healing using a data set of 1,000 patients with 9,000 images. Our long term goal is to investigate the use of these image processing and machine learning models as biomarkers for healing and infection. The objective of this application is to quantify and refine our image segmentation and machine learning models ability to extract healing and infection from patients with diabetes-related foot ulcers treated with total contact casting (TCC) or other by methods. The rationale for this research is that TCC offers the fasted and highest proportion of wounds healed for any modality described in the literature. In this planning study, we will use clinical, laboratory, and image data to refine our models and data collection procedures with our clinical collaborator. Our specific aims are to: 1) Develop image processing models that extract surface area, wound base color and surrounding skin to predict healing and infection in DFU treated with TCC or not; and 2) Analyze predictive ability of image processing model + clinical features to predict healing and infection in DFU treated with TCC or not. Our working hypothesis is image processing and machine learning models will provide valuable prognostic information for healing and infection over changes in surface area alone or clinical features. Our approach is innovative as image processing and machine learning models are relatively new approaches for the wound care field. The proposed research is significant because most wound centers have access to high quality wound images from smartphones or tablets. The prognostic data could help motivate patient adherence and direct more advanced care earlier that could improve patient outcomes.

Progress Reports

Annual Reports

No uploaded documents found.

No uploaded documents found.

No protocols found.

No experiments found.