Unstained White Blood Cell Classification Using Deep Learning

What this paper does

White blood cell morphology is normally read after staining, which costs time and reagents. This study tested whether deep learning can classify unstained cells directly, across 14 classes, on a dataset of 339,885 unstained images from 233 patients split 70/17/13% at patient level. A two-stage pipeline reached mean weighted F1 of 90.33% with a ResNet-18 backbone and 91.77% with a transformer, a 1.44-point gain for the transformer. It is the one piece of my work outside radiology, and it is published as a conference abstract rather than a full paper. I contributed as a co-author.

How

A two-stage classification pipeline over 14 white blood cell classes, comparing a ResNet-18 and a transformer backbone, with a patient-wise 70/17/13% train/validation/test split.

Reported results

• Dataset of 339,885 unstained white blood cell images across 14 classes from 233 patients.
• Mean weighted F1 90.33% with the ResNet-18 backbone and 91.77% with the transformer backbone.
• The transformer gained 1.44 percentage points over ResNet-18 under the patient-wise splitting strategy.

Relevance

If unstained cells can be classified reliably, a preparation step and its reagent cost drop out of the hematology workflow.

What it does not show

Published as a conference abstract in a journal supplement, so methodological detail is limited and the result has not had full-paper peer review.

Citation