Deep learning system for automated white blood cell classification on peripheral blood smears

Source article: Comprehensive Multicenter Evaluation of an AI-Based Deep Learning System for Automated Blood Cell Classification and Morphologic Analysis

Background and objectives Deep learning-based systems have shown promise in automating white blood cell (WBC) classification; however, most validations are limited to single institutions. This study aimed to evaluate the multicenter performance and robustness of a CNN-based deep learning system (DLS) for peripheral blood cell classification and morphologic abnormality detection across three hospitals. Methods A total of 766 peripheral blood smears (613 routine and 153 abnormal) from Juntendo University Hospital,…

By the Truvace deskOctober 16, 20262 min read1 sourcePeer-reviewedTRV-2026-1603· cite & verify

Trace impact readingContested

P 63

Both sides are scored from claims and sources, not community votes.

G 66

In brief

Researchers tested a convolutional neural network system that sorts blood cells on peripheral smears imaged with the Sysmex DI-60. They used 766 smears, 613 routine and 153 abnormal, from Juntendo University Hospital, Kyoto University Hospital, and the National Cancer Center Hospital. The system was trained on 1.49 million annotated images and classifies 16 cell types and 42 morphological features.

The work matters because earlier validations of such tools were mostly confined to single institutions, so consistent results across three labs give stronger evidence for routine use. Agreement was strong for neutrophils, lymphocytes and eosinophils but weaker for monocytes and basophils. The abstract reports performance against manual microscopy, not patient outcomes, so the effect on diagnosis and care is still open.

Main points

The study analyzed 766 peripheral blood smears from three hospitals using the Sysmex DI-60.
The system was trained on 1.49 million annotated cell images and handles 16 cell types and 42 morphological features.
Specificity, negative predictive value and accuracy exceeded 95% across institutions.
Mean F1 scores exceeded 0.80 for major white blood cell types in both routine and abnormal smears.
Monocytes and basophils showed moderate variability against manual counts.

The gain

A CNN-based system analyzing 766 blood smears at three hospitals classified white blood cells with over 95% accuracy and flagged smears with blasts or abnormal cells for expert review.

The problem

Agreement with manual microscopy was only moderate for monocytes and basophils, so the system's counts for these cell types varied more than for other white blood cell types.

The rundown

The sample combined routine and abnormal smears, which tests the system on both everyday and clinically unusual specimens. Of the 766 smears, 613 were routine and 153 were abnormal, and they came from a university hospital, a second university hospital, and a national cancer center.

Performance was measured with standard classification metrics such as accuracy and F1 score. Agreement with manual white blood cell counts was assessed through correlation and regression analyses. For neutrophils, lymphocytes and eosinophils, r2 exceeded 0.8 and Cb exceeded 0.9.

Beyond counting cells, the system flagged smears with blasts or abnormal cells that needed expert review. It also detected morphological features associated with specific diseases. The authors conclude this supports clinical applicability for routine differentials and disease-specific abnormality detection.

What this doesn’t fix

Agreement with manual microscopy was only moderate for monocytes and basophils, so performance is not uniform across all cell types, and the abstract reports no patient outcome data.