Burn wound classification model using spatial frequency-domain imaging and machine learning | JoVE Visualize

Area of Science:

Biomedical Optics
Medical Imaging
Machine Learning in Medicine

Background:

Accurate burn severity assessment is crucial for effective wound care and treatment, as delays increase risks of scarring and infection.
Existing imaging techniques and machine learning models have been used to infer burn severity, but limitations remain in early and accurate classification.
Spatial Frequency-Domain Imaging (SFDI) offers a non-invasive method to characterize tissue optical properties, showing potential for burn assessment.

Purpose of the Study:

To investigate the feasibility of using SFDI reflectance data with a support vector machine (SVM) classifier to predict burn severity.
To evaluate the accuracy of different SFDI data subsets (wavelengths, spatial frequencies, relative values) for burn severity prediction in a porcine model.

Main Methods:

Calibrated reflectance images were acquired using SFDI across eight wavelengths (471-851 nm) and five spatial frequencies (0-0.2 mm⁻¹) in a porcine burn model.
Three cubic SVM models were trained and tested using distinct subsets of SFDI data: planar illumination only, all wavelengths and spatial frequencies, and data relative to unburned tissue.
Model accuracy was validated against 28-day burn status using leave-one-out cross-validation.

Main Results:

The SVM model utilizing all wavelengths and spatial frequencies achieved 92.5% accuracy in predicting burn severity at 24 hours post-injury.
The model using SFDI data relative to unburned tissue demonstrated the highest accuracy at 94.4%.
A model using only planar illumination (0 mm⁻¹) achieved 88.8% accuracy, highlighting the benefit of multi-frequency data.

Conclusions:

The combination of SFDI and machine learning, particularly SVM, shows significant potential for accurate and early burn severity prediction.
Utilizing multiple spatial frequencies and wavelengths, or data relative to unburned tissue, enhances predictive accuracy.
This approach could lead to improved burn management, reducing complications like scarring and infection.