Photo Credit: MicrovOne

The following is a summary of “A proof of concept for microcirculation monitoring using machine learning based hyperspectral imaging in critically ill patients: a monocentric observational study,” published in the July 2024 issue of Critical Care by Kohnke et al.

Sepsis development involves impaired microcirculation affecting tissue oxygenation, with hyperspectral imaging (HSI) and machine learning (ML) potentially enabling automated bedside assessment of tissue characteristics in the ICU.

Researchers conducted a retrospective study assessing MLs’ ability to identify specific hand regions, differentiate between HCs, patients with critical illness, and sepsis using HSI.

They recorded HSI raw data from 75 patients with critical illness and sepsis and 30 HCs using the TIVITA® Tissue System and analyzed the data with an automated ML approach. SOFA score was used to divide patients into 2 groups, less severely ill (SOFA ≤ 5) and severely ill (SOFA > 5), for further subanalysis. MediaPipe was used for fully automated ROI detection (palm and fingertips) and feature extraction. Statistical analysis was performed using the Mann–Whitney-U test with Benjamini, Krieger, and Yekutieli (BKY) correction to highlight relevant wavelength combinations. Additionally, a trained random forest model using bootstrapping was used to determine the importance of features in understanding the wavelength significance in model decision-making.

The result showed the successful establishment of an automated pipeline for generating ROIs and extracting HSI features. The analysis of HSI raw data effectively differentiated between HCs and patients with sepsis. Fingertips exhibited significant differences in the 575–695 nm and 840–1000 nm wavelength ranges, while the palm showed significant differences in the 925–1000 nm range. Feature importance plots highlighted relevant information within the same wavelength ranges. The combined analysis of palm and fingertip data yielded the highest reliability, achieving an AUC of 0.92 in distinguishing patients with sepsis from HCs.

Investigators concluded that the integration of automated and standardized ROIs with skin HSI analysis reliably differentiated between HCs and patients with sepsis, enabling rapid and objective assessment of skin microcirculation in patients with critical illness.

Source: ccforum.biomedcentral.com/articles/10.1186/s13054-024-05023-w