A new PHOREVER publication entitled “A feasible machine learning framework for diagnosing stroke patients versus mimic conditions incorporating extracellular vesicle characterization and EHR features fusion” presents promising results on improving stroke diagnosis through the fusion of extracellular vesicle (EV) analysis and electronic health record (EHR) data using machine learning techniques.
The study addresses a critical challenge in emergency medicine: the rapid and accurate differentiation between stroke and stroke-mimicking conditions. By combining demographic, clinical, and biochemical patient data with EV features extracted from blood plasma using flow cytometry, the proposed artificial intelligence–driven decision support system enables a minimally invasive and low-resource diagnostic approach.
Blood samples from 140 patients with stroke or stroke-mimicking conditions were analyzed. Statistical characteristics of EV size distribution and concentration were integrated with standard EHR variables and evaluated using a decision tree learning model. The resulting framework achieved a weighted classification accuracy of 86%, demonstrating strong potential to support clinical decision-making in time-critical emergency settings.
These findings highlight how photonics-based EV characterization, combined with machine learning, can enhance diagnostic accuracy and aid treatment strategies in stroke care.
The publication is authored by V. E. Katsigiannis, I. Kakkos, S. T. Miloulis, I. A. Vezakis, O. Petropoulou, J. Rops, N. C. Buntsma, E. van der Pol, D. I. Fotiadis and G. K. Matsopoulos.
Read the full publication here.