Hyperspectral imaging in the emergency department to characterize lower leg edema
- 1Laboratory of Sensorimotor Research, National Eye Institute, NIH, Bethesda MD 20892
Abstract
Interactions between light and biological tissue could reveal disease-related changes that could be advantageous for rapid diagnosis when applied to whole limbs. Here, we tested the extent to which visible/near-infra-red hyperspectral imaging (HSI) deployed in the emergency department could classify cellulitis, non-cellulitis edema, and healthy lower-leg tissue across skin pigmentation types. We collected HSI data of the lower leg from 83 emergency-room patients and analyzed the calibrated normalized spectra (400-1000 nm) using a machine-learning classification algorithm, evaluated by cross-validation and benchmarked against standard spectral indices (oxygenation, hemoglobin, water, near-infrared perfusion). Classification at patient-level and pixel-level resolution performed comparably well using whole-spectrum and indices, for cellulitis versus healthy (AUC: 0.93 vs. 0.96) and non-cellulitis edema versus cellulitis (AUC: 0.92 vs. 0.89). Full-spectrum analysis substantially improved non-cellulitis edema versus healthy classification (patient-level AUC: 0.72 vs. 0.41; pixel-level AUC: 0.78 vs. 0.50), with no significant effect of skin type on accuracy.
Related
- Integrating structured and unstructured EHR data to characterize social and behavioral factors associated with frequent emergency department use among patients with cardiovascular disease
- Ethnoracial disparities in emergency department triage are more pronounced in lower-risk visits: a multicenter study
- Emergency Department Overcrowding Despite Lower Post-Pandemic Volumes: A Multicenter Study
- Preoperative GLP-1 receptor agonist use is associated with lower 2-year emergency department utilization and all-cause mortality after total shoulder arthroplasty
- Orthopaedic Consultation Versus No Consultation for Clavicle, Radial Head, and Proximal Humerus Fractures: Association with Post-Imaging Emergency Department Length of Stay
BibTeX
@misc{kasi2026hyperspectral,
title = {{Hyperspectral imaging in the emergency department to characterize lower leg edema}},
author = {K. Kasi and C. May and H. Gallegos and L. Kobayashi and B. R. Conway and J. R. Pare},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.15.26363173},
url = {https://www.medrxiv.org/content/10.64898/2026.09.15.26363173v1},
note = {CC0},
}
RIS
TY - UNPB
TI - Hyperspectral imaging in the emergency department to characterize lower leg edema
AU - K. Kasi
AU - C. May
AU - H. Gallegos
AU - L. Kobayashi
AU - B. R. Conway
AU - J. R. Pare
PY - 2026
DA - 2026/09/15
PB - medRxiv
DO - 10.64898/2026.09.15.26363173
UR - https://www.medrxiv.org/content/10.64898/2026.09.15.26363173v1
AB - Interactions between light and biological tissue could reveal disease-related changes that could be advantageous for rapid diagnosis when applied to whole limbs. Here, we tested the extent to which visible/near-infra-red hyperspectral imaging (HSI) deployed in the emergency department could classify cellulitis, non-cellulitis edema, and healthy lower-leg tissue across skin pigmentation types. We collected HSI data of the lower leg from 83 emergency-room patients and analyzed the calibrated normalized spectra (400-1000 nm) using a machine-learning classification algorithm, evaluated by cross-validation and benchmarked against standard spectral indices (oxygenation, hemoglobin, water, near-infrared perfusion). Classification at patient-level and pixel-level resolution performed comparably well using whole-spectrum and indices, for cellulitis versus healthy (AUC: 0.93 vs. 0.96) and non-cellulitis edema versus cellulitis (AUC: 0.92 vs. 0.89). Full-spectrum analysis substantially improved non-cellulitis edema versus healthy classification (patient-level AUC: 0.72 vs. 0.41; pixel-level AUC: 0.78 vs. 0.50), with no significant effect of skin type on accuracy.
ER -