Machine-Learning-Assisted Procoagulant Extracellular Vesicle Barcode Assay toward High-Performance Evaluation of Thrombosis-Induced Death Risk in Cancer Patients.

Abstract

Venous thromboembolism (VTE) is the most fatal complication in cancer patients. Unfortunately, the frequent misdiagnosis of VTE owing to the lack of accurate and efficient evaluation approaches may cause belated medical intervention and even sudden death. Herein, we present a rapid, easily operable, highly specific, and highly sensitive procoagulant extracellular vesicle barcode (PEVB) assay composed of TiO₂ nanoflower (TiNFs) for visually evaluating VTE risk in cancer patients. TiNFs demonstrate rapid label-free EV capture capability by the synergetic effect of TiO₂-phospholipids molecular interactions and topological interactions between TiNFs and EVs. From ordinary plasma samples, the PEVB assay can evaluate potential VTE risk by integrating TiNFs-based EV capture and in situ EV procoagulant ability test with machine-learning-assisted clinical data analysis. We demonstrate the feasibility of this PEVB assay in VTE risk evaluation by screening 167 cancer patients, as well as the high specificity (97.1%) and high sensitivity (96.8%), fully exceeding the nonspecific and posterior traditional VTE test. Together, we proposed a TiNFs platform allowing for highly accurate and timely diagnosis of VTE in cancer patients.