Summary
A medRxiv preprint reports that label-free imaging of 10 μL whole-blood samples distinguished sepsis from other critical illness and healthy controls in a prospective study. The study included 29 critically ill adults, so its performance findings are preliminary.
A medRxiv preprint reports that Remoscope, a label-free whole-blood imaging system, distinguished sepsis from non-infectious critical illness and healthy controls in a prospective study. The researchers analysed 10 μL blood samples from 29 critically ill adults and 22 healthy controls. Their model, optimised to rule out sepsis, had reported sensitivity of 100% and specificity of 85.3% in the study. These early results come from a small patient cohort and are not yet a clinical validation.
How the imaging approach works
The system uses imaging cytometry to capture blood-cell images without staining the sample. The researchers used a multiple-instance-learning (MIL) model to combine features from thousands of images for each patient. In this approach, the model works with collections of image examples rather than treating a single image as the whole sample.
The study’s authors report that model heatmaps, unsupervised groupings of individual cells and measurements of cell shape all pointed to changes in white and red blood cells. These included immature neutrophil-like cells. The findings suggest that blood-cell morphology—the cells’ visible form and structure—may carry signals associated with sepsis.
The authors describe the method as requiring neither stains nor a cold chain. If its performance is confirmed in further clinical studies, that combination could simplify sample handling. The preprint does not give a specific time from sampling to result, so the word “rapid” should not be read as a measured turnaround time in this report.
What the study measured
Multiple physicians classified the 29 critically ill adults: 13 had sepsis, 12 had non-infectious critical illness and four had indeterminate status. The researchers also compared the model’s results with samples from 22 healthy controls. They report that the three-class model detected sepsis with 100% sensitivity and 85.3% specificity, and say it was optimised for rule-out. Sensitivity and specificity describe how often a test correctly identifies people with and without the condition, respectively; the reported figures are estimates from this study, not a guarantee of performance in other hospitals or patient groups.
The work is posted as a preprint, rather than a peer-reviewed journal article. The cohort included only 13 patients classified as having sepsis, making the performance estimates preliminary. The abstract does not describe an independent external validation cohort.
The authors also disclosed relevant commercial interests: Paul Lebel is founder and CEO of Oxalis Imaging, and Nicholas Bratvold, Ilakkiyan Jeyakumar and Gregory Suematsu are company founders. The authors are named inventors on provisional patent application No. 63/957,084, related to the technology.