top of page

New Publication(Deep-Learning Analysis of Peripheral Blood Smear for Prediction of Acute Inflammatory States and Chronic Comorbidities) in JMIR Medical Informatics

6시간 전
1분 분량

We are pleased to announce that our team has published a new paper in JMIR Medical Informatics (Impact Factor: 5.0), Vol. 14, November 2026.

This study develops and internally validates an integrated deep learning framework that uses cell-level morphology from routine peripheral blood smear (PBS) examinations to discriminate acute inflammatory states and chronic comorbidities. Although PBS examination reveals immune cell morphology that reflects systemic pathophysiology, its predictive value for systemic states has not been systematically demonstrated. The proposed framework extracts cell-level morphological features and aggregates them into slide-level representations, using 51,369 PBS examinations for self-supervised cell-level pretraining and 9,211 examinations for slide-level downstream tasks. For acute states—white blood cell count, elevated CRP, elevated ESR, and septic shock—the model achieved AUROCs of approximately 0.84–0.89 across the development test and holdout sets, while discrimination for chronic comorbidities (Charlson Comorbidity Index, cardiovascular disease, chronic kidney disease, and diabetes mellitus) was more modest. The framework significantly outperformed a cell-type-only model for all outcomes in both datasets, demonstrating its potential as a tool for automated PBS analysis.

“Deep-Learning Analysis of Peripheral Blood Smear for Prediction of Acute Inflammatory States and Chronic Comorbidities: Development and Internal Validation Study” Jaehyeok Jang, Minsu Lee, Yu Rang Park, & Jaewoo Song (2026). JMIR Medical Informatics, 14, e100445.

 
 
 

댓글


연세대 의과대학.PNG

​

© 2019 DHLab, All Rights Reserved

bottom of page