Abstract
With sepsis remaining a leading cause of mortality, early identification of septic patients and those at high risk of death is a challenge of high socioeconomic importance. The driving hypothesis of this study was that hyperspectral imaging (HSI) could provide novel biomarkers for sepsis diagnosis and treatment management due to its potential to monitor microcirculatory alterations. We conducted a comprehensive study involving HSI data of the palm and fingers from more than 480 patients on the day of their intensive care unit (ICU) admission. The findings demonstrate that HSI measurements can predict sepsis with an area under the receiver operating characteristic curve (AUROC) of 0.80 (95 % confidence interval (CI) [0.76; 0.84]) and mortality with an AUROC of 0.72 (95 % CI [0.65; 0.79]). The predictive performance improves substantially when additional clinical data is incorporated, leading to an AUROC of up to 0.94 (95 % CI [0.92; 0.96]) for sepsis and 0.84 (95 % CI [0.78; 0.89]) for mortality. We conclude that HSI presents novel imaging biomarkers for the rapid, non-invasive prediction of sepsis and mortality, suggesting its potential as an important modality for guiding diagnosis and treatment.
Introduction
Sepsis is a leading cause of death worldwide, yet early diagnosis remains a significant clinical hurdle. Current protocols often rely on identifying organ dysfunction—which may only manifest after substantial damage—or non-specific biomarkers. A critical challenge in the Intensive Care Unit (ICU) is distinguishing between patients who are critically ill for non-septic reasons (such as major surgery or trauma) and those developing sepsis. This study addresses this "real-world" challenge by identifying sepsis within a cohort where all patients are already severely ill, avoiding the pitfalls of "shortcut learning" common in studies comparing septic patients to healthy volunteers.
Methods and Materials
The researchers utilized Hyperspectral Imaging (HSI), a non-invasive technique that captures 100 spectral channels (500–1000 nm) per pixel. This allows for the measurement of functional tissue parameters—such as oxygen saturation, hemoglobin levels, and water content—which serve as indicators of microcirculatory health.
The Pipeline
- Data Acquisition: A 7-second scan of the palm and fingers was performed using a medical-grade HSI system (Tivita® 2.0).
- Deep Learning: A ResNet14d convolutional neural network architecture was employed to process the 3D HSI cubes.
- Multimodal Fusion: A late-fusion approach combined HSI data with structured clinical data (vitals, lab results, and medication doses) to enhance predictive accuracy.
Datasets
The study utilized a prospective observational dataset of over 480 adult patients admitted to a surgical ICU in Germany. The sepsis diagnosis cohort included 437 patients (129 septic), while the mortality prediction cohort included 483 patients (68 non-survivors within 30 days). Clinical data encompassed 45 parameters, including demographics and blood gas analysis.
Results
The study revealed that imaging the palm provided superior results compared to the finger, likely due to lower melanin content and better accessibility. Septic patients and non-survivors exhibited a distinct physiological profile: significantly lower tissue oxygen saturation coupled with higher tissue water (edema) and hemoglobin content.
Key performance metrics included:
- Sepsis Diagnosis: HSI of the palm alone achieved an AUROC of 0.80. When combined with clinical data, this rose to 0.94 AUROC.
- Mortality Prediction: HSI of the palm alone achieved an AUROC of 0.72, increasing to 0.84 when integrated with clinical data.
- Clinical Comparison: The AI models consistently outperformed standard clinical tools such as the qSOFA score, SIRS criteria, and traditional biomarkers like Procalcitonin (PCT) and C-reactive protein (CRP).
Notably, adding just one clinical parameter—the dose of noradrenaline—boosted the sepsis diagnosis AUROC from 0.80 to 0.87, demonstrating the power of multimodal assessment.
Discussion and Limitations
HSI presents a rapid, bedside assessment tool that provides results in seconds, offering a significant speed advantage over lab-based biomarkers. However, the study was conducted at a single center with a focus on abdominal infections, necessitating further validation across diverse patient populations and infection types. Future research will explore longitudinal monitoring to track treatment response in real-time and the development of more compact, low-cost HSI sensors for broader clinical use.
Summary updated