Use of Artificial Intelligence in Burn Assessment: A Scoping Review with a Large Language Model-Generated Decision...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Burn Surgery

Background:

• Burns result in significant global mortality and morbidity, particularly in LMICs.
• Current burn depth and TBSA assessment is subjective, relying on visual and bedside examination.
• Convolutional neural networks (CNNs) offer potential for objective, image-based burn assessment, but clinical applicability is unproven.

Purpose of the Study:

• To review evidence on CNN performance for burn TBSA, depth, and treatment tasks.
• To assess the feasibility of using a large language model (LLM) to create a decision tree from extracted findings.

Main Methods:

• A scoping review following PRISMA-ScR guidelines.
• Searched PubMed, Web of Science, and Cochrane databases.
• Included 24 studies analyzing 2D burn images with CNNs and reporting performance metrics.

Main Results:

• Reported CNN performance for TBSA and depth assessment was often high but highly variable.
• Significant heterogeneity existed in study designs, datasets, imaging, and validation methods.
• A single study showed high accuracy for graft prediction, but results are not generalizable.

Conclusions:

• CNNs demonstrate potential for burn TBSA and depth evaluation.
• Heterogeneity, data limitations, and insufficient external validation impede clinical translation.
• The LLM-generated decision tree serves as a literature synthesis tool, not a clinical decision support.