A proof of concept for microcirculation monitoring using machine learning based hyperspectral imaging in critically ill patients: a monocentric observational study

Abstract

Background: Impaired microcirculation is a cornerstone of sepsis development and leads to reduced tissue oxygenation, influenced by fluid and catecholamine administration during treatment. Hyperspectral imaging (HSI) is a non-invasive bedside technology for visualizing physicochemical tissue characteristics. Machine learning (ML) for skin HSI might offer an automated approach for bedside microcirculation assessment, providing an individualized tissue fingerprint of critically ill patients in intensive care. The study aimed to determine if machine learning could be utilized to automatically identify regions of interest (ROIs) in the hand, thereby distinguishing between healthy individuals and critically ill patients with sepsis using HSI.

Methods: HSI raw data from 75 critically ill sepsis patients and from 30 healthy controls were recorded using TIVITA® Tissue System and analyzed using an automated ML approach. Additionally, patients were divided into two groups based on their SOFA scores for further subanalysis: less severely ill (SOFA ≤ 5) and severely ill (SOFA > 5). The analysis of the HSI raw data was fully-automated using MediaPipe for ROI detection (palm and fingertips) and feature extraction. HSI Features were statistically analyzed to highlight relevant wavelength combinations using Mann-Whitney-U test and Benjamini, Krieger, and Yekutieli (BKY) correction. In addition, Random Forest models were trained using bootstrapping, and feature importances were determined to gain insights regarding the wavelength importance for a model decision.

Results: An automated pipeline for generating ROIs and HSI feature extraction was successfully established. HSI raw data analysis accurately distinguished healthy controls from sepsis patients. Wavelengths at the fingertips differed in the ranges of 575-695 nm and 840-1000 nm. For the palm, significant differences were observed in the range of 925-1000 nm. Feature importance plots indicated relevant information in the same wavelength ranges. Combining palm and fingertip analysis provided the highest reliability, with an AUC of 0.92 to distinguish between sepsis patients and healthy controls.

Conclusion: Based on this proof of concept, the integration of automated and standardized ROIs along with automated skin HSI analyzes, was able to differentiate between healthy individuals and patients with sepsis. This approach offers a reliable and objective assessment of skin microcirculation, facilitating the rapid identification of critically ill patients.

Fig. 1

HSI measurements of the inside of the patient's hand are carried out within 24 h after admission to ICU. At the same time, demographic data, clinical data and the SOFA score are recorded in order to assess the patient's state of health. Automatic ROI detection and a machine learning-based analysis of the raw HSI data were performed

Fig. 2

A Overview of the methodical procedure. Landmarks are detected on RGB hand images as reference points using the MediaPipe network. These reference points are utilized to generate ROI on the hand. The positions of the ROIs are transferred into the HSI cube, and the average spectrum of pixels is computed. This spectrum can be analyzed and utilized for machine learning inquiries. B Examples of hands with highlighted ROIs for palm and fingertips across different patients

Fig. 3

Each ROI covers wavelengths from 500 to 1000 nm with a spectral resolution of 5 nm, resulting in 100 measured wavelengths. The combination of the two ROIs results in a spectrum of 200 wavelengths

Fig. 4

Mean Spectra with 95% Confidence intervals for control and for septic patients. Comparison of the spectra from healthy control (dark blue) and patients with sepsis (pink) for two distinct ROI: palm (left) and fingertips (right). In the middle, the difference between sepsis and control for each wavelength is shown. The colors in the difference plots are chosen such that the color indicates which spectrum (with the same color) exhibits higher absorption. Bar plots positioned below highlight significant differences between the spectral distributions of the groups, showcasing the discriminative spectral bands. Differences were determined to be statistically significant using the Mann–Whitney-U Test, with a p-value below 0.05. Test correction was performed using the BKY approach

Fig. 5

Mean Spectra Comparison with 95% Confidence Intervals. Comparison of the spectra for the palm and fingertips across three distinct patient groups: SOFA-score: healthy control (black), less severe organ dysfunction = SOFA ≤ 5 (dark blue) and severe organ dysfunction = SOFA > 5 (red). Below the spectra, difference plots between the patients with sepsis and the healthy controls are shown as bar plots, as well as bar plots highlight significant differences between the spectral distributions of the groups, showcasing the discriminative spectral bands. Differences were determined to be statistically significant using the Mann–Whitney-U Test, with a p-value below 0.05. Test correction was performed using the BKY approach. The colors in the difference plots are chosen such that the color indicates which spectrum (with the same color) exhibits higher absorption

Fig. 6

Classification results of healthy control and septic patients for palm (dark blue), fingertips (grey) and combination of both (pink) ROI. The ROC curves of the three ROIs are shown on the left, and on the right are the feature importance plots showing which wavelengths were relevant for the classification. For the combined ROI, the mean feature importance of palm wavelengths and fingertips wavelengths was calculated