Abstract
Nailfold capillaroscopy is a standard clinical tool for visualizing structural microvascular abnormalities in patients with systemic sclerosis (SSc). However, structural changes often lag behind functional physiological shifts. The CAPoxy study investigated whether multispectral imaging (MSI) could provide a functional assessment of capillaries—specifically regarding hemoglobin concentration and oxygenation—to improve discrimination between healthy controls and patients with SSc. Using a nine-band multispectral camera on a cohort of 22 individuals, the study identified significant spectral differences in the tissue surrounding capillaries. A machine learning model trained on this data achieved 84% accuracy in classifying SSc, suggesting that MSI can detect microvascular distress before or alongside structural deformation.
Introduction
Systemic sclerosis (SSc) is a chronic autoimmune condition characterized by progressive microvascular damage and tissue hypoxia. While traditional nailfold capillaroscopy allows clinicians to observe structural hallmarks like giant capillaries or hemorrhages, these markers represent late-stage architectural changes. There is a clinical need for "functional" imaging that can measure real-time physiological states, such as blood oxygenation, which may provide earlier diagnostic signals.
While pulse oximetry measures systemic arterial oxygenation, it lacks the spatial resolution to assess the specific microenvironment of the nailfold. Multispectral imaging (MSI) addresses this gap by capturing data across multiple narrow wavelength bands, allowing for the visualization of hemoglobin distribution and oxygen saturation gradients at the capillary level.
Methodology
Hardware and Data Acquisition
The researchers developed a custom MSI capillaroscopy system utilizing a 9-band multispectral camera (CMS-V, SILIOS). This sensor captures distinct wavelength bands between 553 nm and 829 nm. The system was integrated with a white LED ring light and a magnifying lens to focus on the nailfold bed.
Image Processing and Analysis
Raw data underwent a demosaicking process to reconstruct full-resolution images for each spectral band. To account for variations in light intensity and background noise, a correction procedure was applied using reflectance standards.
The analysis focused on three specific Regions of Interest (ROIs):
- Capillaries: The interior of the visible vessels.
- Edges: The tissue within a 20-pixel radius of the capillary wall.
- Outside: The surrounding interstitial tissue.
Ratiometric and Machine Learning Approaches
To highlight oxygenation differences, the team employed ratiometric imaging (e.g., the 587 nm / 553 nm ratio). Because oxygenated and deoxygenated hemoglobin absorb light differently across these wavelengths, these ratios serve as proxies for oxygen saturation. Finally, a Random Forest (RF) classifier was trained on the spectral signatures to evaluate the system's diagnostic potential.
Results and Discussion
The Perivascular "Invisible" Signature
A primary finding of the CAPoxy study was that SSc-related changes are not confined to the vessels themselves. Significant spectral differences (p < 0.001) were observed in the tissue surrounding the capillaries. This suggests that MSI can detect physiological stress in the microenvironment even when the capillary structure appears relatively intact.
Hemoglobin and Oxygenation Metrics
Patients with SSc demonstrated higher indicators of hemoglobin concentration in the regions immediately adjacent to capillaries compared to healthy controls. Furthermore, the 587/553 nm ratio—an indicator of lower oxygenation—was 5% higher in the SSc group. This supports the hypothesis that SSc patients exist in a state of chronic local hypoxia.
Classification Performance
The Random Forest model successfully distinguished SSc patients from healthy controls with an accuracy of 84%. This high level of performance in a feasibility study underscores the potential for MSI to serve as an automated diagnostic aid.
Limitations and Future Directions
As a feasibility study, the sample size was limited to 22 participants (with 16 providing high-quality data for final analysis). Future research will require larger, multi-center cohorts to validate these findings across different skin tones and disease stages.
Technical improvements are also planned, including the use of halogen light sources to improve signal-to-noise ratios at longer wavelengths (>714 nm) and expanding the spectral range to 400–550 nm. Such expansions would allow for "linear spectral unmixing," moving from weighted ratios to absolute measurements of oxygen saturation.
Conclusion
The CAPoxy study demonstrates that multispectral imaging adds a vital functional layer to nailfold capillaroscopy. By detecting hemoglobin and oxygenation variances in the perivascular tissue, MSI offers a promising path toward earlier detection and more nuanced monitoring of microvascular diseases like systemic sclerosis.
Summary updated