A Feasibility Study of Thermography for Detecting Pressure Injuries Across Diverse Skin Tones

Miriam Asare-Baiden,Sharon Eve Sonenblum,Kathleen Jordan,Andrew Chung,Judy W Gicchoya,Vicki Stover Hertzberg,Joyce C Ho

DOI: https://doi.org/10.1101/2024.10.14.24315465

2024-10-16

Abstract:Pressure injury (PI) detection is challenging, especially in dark skin tones, due to the unreliability of visual inspection. Thermography may serve as a viable alternative as temperature differences in the skin can indicate impending tissue damage. Although deep learning models hold considerable promise toward reliably detecting PI, existing work fails to evaluate performance on diverse skin tones and varying data collection protocols. We collected a new dataset of 35 participants focused on darker skin tones where temperature differences are induced through cooling and cupping protocols. The dataset includes different cameras, lighting, patient pose, and camera distance. We compare the performance of three convolutional neural network (CNN) models trained on either the thermal or the optical images on all skin tones. Our results suggest thermography-based CNN is robust to data collection protocols. Moreover, the visual explanation often captures the region of interest without requiring explicit bounding box labels.

Health Informatics

What problem does this paper attempt to address?

This paper aims to address the challenges faced by pressure injury (PI) detection in individuals with dark skin. Specifically, existing visual inspection methods have difficulty in reliably identifying PI on dark skin, which has led to ethnic disparities and poor PI prognoses. The paper proposes the use of thermography as an alternative because changes in skin temperature can indicate impending tissue damage. Moreover, although deep - learning models such as convolutional neural networks (CNNs) have shown high predictive performance in PI detection, existing studies have failed to evaluate the performance of these models under different skin colors and data - collection protocols.

### Main research objectives:
1. **Verify the effectiveness of thermography on different skin colors**: Specifically for individuals with dark skin, study whether thermography can effectively detect temperature changes.
2. **Compare the effectiveness of thermography and optical imaging**: Evaluate the performance differences between thermography and traditional optical imaging in PI detection.
3. **Evaluate the impact of data - collection protocols**: Study the effects of different lighting conditions, patient postures, camera distances, and camera resolutions on model performance.
4. **Analyze the interpretability of the model**: Through visualization techniques such as Grad - CAM, evaluate whether the model can automatically identify regions of interest without explicitly labeling bounding boxes.

### Research methods:
- **Data collection**: Collect optical and thermographic images from 35 healthy adults (mostly with dark skin), and induce temperature changes through two protocols: cooling and cupping.
- **Model selection**: Use three CNN models (MobileNetV2, ResNet50, InceptionNetV3) for classification tasks.
- **Task classification**: Evaluate the performance of three image types (optical images, black - and - white thermography, color thermography) in two tasks: cooling and erythema.
- **Performance evaluation**: Use metrics such as accuracy, sensitivity, specificity, F1 - score, and AUC to evaluate model performance.

### Main findings:
- **Thermography is superior to optical imaging**: In the cooling task, thermography (both black - and - white and color) showed perfect performance in all three models, while in optical imaging, the performance was significantly lower.
- **Thermography is insensitive to data - collection protocols**: Thermography models can still maintain high performance under different lighting conditions, patient postures, and camera distances.
- **The model has good interpretability**: Through Grad - CAM visualization, the model can correctly identify cooling and erythema regions without explicitly labeling bounding boxes.

### Conclusion:
This study demonstrates the potential of thermography combined with CNN models in detecting PI in individuals with dark skin, especially in different clinical environments and data - collection conditions. These findings provide strong support for the future adoption of thermography in clinical practice. However, the study also points out limitations such as small sample size and experimental - environment control, and further research is needed to verify these preliminary results.