Samantha Huang1, Justin Dang2, Clifford C Sheckter3, Haig A Yenikomshian4, Justin Gillenwater5. 1. Keck School of Medicine, University of Southern California, Los Angeles, CA, United States; Los Angeles County Regional Burn Center, Los Angeles County + University of Southern California Medical Center, Los Angeles, CA, United States. 2. Los Angeles County Regional Burn Center, Los Angeles County + University of Southern California Medical Center, Los Angeles, CA, United States. 3. Northern California Regional Burn Center at Santa Clara Valley Medical Center, Division of Plastic & Reconstructive Surgery, Stanford University, Stanford, CA, United States. 4. Division of Plastic and Reconstructive Surgery, Keck School of Medicine, University of Southern California, Los Angeles, CA, United States; Los Angeles County Regional Burn Center, Los Angeles County + University of Southern California Medical Center, Los Angeles, CA, United States. 5. Division of Plastic and Reconstructive Surgery, Keck School of Medicine, University of Southern California, Los Angeles, CA, United States; Los Angeles County Regional Burn Center, Los Angeles County + University of Southern California Medical Center, Los Angeles, CA, United States. Electronic address: Justin.Gillenwater@med.usc.edu.
Abstract
BACKGROUND: Visual evaluation is the most common method of evaluating burn wounds. Its subjective nature can lead to inaccurate diagnoses and inappropriate burn center referrals. Machine learning may provide an objective solution. The objective of this study is to summarize the literature on ML in burn wound evaluation. METHODS: A systematic review of articles published between January 2000 and January 2021 was performed using PubMed and MEDLINE (OVID). Articles reporting on ML or automation to evaluate burn wounds were included. Keywords included burns, machine/deep learning, artificial intelligence, burn classification technology, and mobile applications. Data were extracted on study design, method of data acquisition, machine learning techniques, and machine learning accuracy. RESULTS: Thirty articles were included. Nine studies used machine learning and automation to estimate percent total body surface area (%TBSA) burned, 4 calculated fluid estimations, 19 estimated burn depth, 5 estimated need for surgery, and 2 evaluated scarring. Models calculating %TBSA burned demonstrated accuracies comparable to or better than paper methods. Burn depth classification models achieved accuracies of >83%. CONCLUSION: Machine learning provides an objective adjunct that may improve diagnostic accuracy in evaluating burn wound severity. Existing models remain in the early stages with future studies needed to assess their clinical feasibility.
BACKGROUND: Visual evaluation is the most common method of evaluating burn wounds. Its subjective nature can lead to inaccurate diagnoses and inappropriate burn center referrals. Machine learning may provide an objective solution. The objective of this study is to summarize the literature on ML in burn wound evaluation. METHODS: A systematic review of articles published between January 2000 and January 2021 was performed using PubMed and MEDLINE (OVID). Articles reporting on ML or automation to evaluate burn wounds were included. Keywords included burns, machine/deep learning, artificial intelligence, burn classification technology, and mobile applications. Data were extracted on study design, method of data acquisition, machine learning techniques, and machine learning accuracy. RESULTS: Thirty articles were included. Nine studies used machine learning and automation to estimate percent total body surface area (%TBSA) burned, 4 calculated fluid estimations, 19 estimated burn depth, 5 estimated need for surgery, and 2 evaluated scarring. Models calculating %TBSA burned demonstrated accuracies comparable to or better than paper methods. Burn depth classification models achieved accuracies of >83%. CONCLUSION: Machine learning provides an objective adjunct that may improve diagnostic accuracy in evaluating burn wound severity. Existing models remain in the early stages with future studies needed to assess their clinical feasibility.