Find this paper

Abstract

Sepsis is life-threatening, yet diagnosis remains a major global challenge because existing approaches lack speed and accuracy. Here we show that label-free imaging cytometry of 10 {micro}L whole-blood samples by Remoscope rapidly distinguishes patients with sepsis from those with non-infectious critical illness and healthy controls. In a prospective study of 29 critically ill adults, multi-physician adjudication identified sepsis (n=13), non-infectious critical illness (n=12), and indeterminate status (n=4); these patients were compared to healthy controls (n=22). A multiple instance learning (MIL) model pooled unbiased features from thousands of blood images per patient. Optimized for rule-out, the three-class model detected sepsis with 100% sensitivity and 85.3% specificity. Class activation heatmaps, unsupervised single cell embeddings, and morphometrics converged on abnormal leukocyte and erythrocyte morphology, including immature neutrophil-like cells. Label-free whole-blood imaging with MIL detects sepsis-associated hematologic signatures without stains or cold chain, establishing a foundation for rapid critical care diagnostics.

Posts about this paper

No posts about this paper yet. Mention it with @ in a post to start the conversation.