Two-step regression analysis approach to assess burn wound severity using spatial frequency domain imaging

Prompt care is essential for burn wound recovery. Spatial frequency domain imaging (SFDI) has previously shown promise in predicting healing outcomes across burn severity grades. This study builds on that by demonstrating calibrated reflectance images ( $R_d$ ) from SFDI can estimate thermally induced collagen denaturation depth (CDD), a histology-based metric of burn severity linked to healing outcomes. These findings may simplify future hardware design by clarifying contrast sources in SFDI.

To develop predictive models for: 1) Day-1 postburn CDD using SFDI $R_d$ and 2) Day-28 healing outcomes using day-1 CDD.

Using a previously reported graded-severity porcine burn model ( $n=4$ ) with eight contact durations (5 to 40 s), we collected SFDI and color images on days 0, 1, 3, 7, 14, 21, and 28. Histological analysis using Picrosirius red staining and polarization microscopy was performed on days 1, 7, 14, 21, and 28 to assess CDD. Healing outcomes were clinically evaluated on day 28. For analysis, a two-step regression framework was applied:Step 1: Multiple linear regression, where day-1 SFDI R_d is used to predict same-day CDD.Step 2: Logistic regression, where day-1 CDD is used to predict day-28 healing outcome.Together, these steps established a regression framework to predict day-1 CDD and day-28 healing outcomes using day-1 SFDI R_d.

The linear model using $R_d$ across eight wavelengths (471-851 nm) and five spatial frequencies (0 to $0.2{\mathrm{mm}}^{-1}$ ) predicted CDD with a root mean square error of $105\mu m$ and adjusted $R^2$ of 0.71. The logistic model predicted healing outcomes with an ROC-AUC of 0.88, supporting CDD as an early indicator for burn severity assessed by healing potential.

This two-step framework enables early prediction (as early as day 1) of burn severity and healing using SFDI $R_d$ .