Laboratory Investigation · 2020
Machine-based Detection and Classification for Bone Marrow Aspirate Differential Counts
A two-stage computer-vision system for detecting and classifying white blood cells in bone-marrow aspirate images.
- Digital pathology
- Cell detection
- Medical imaging
Research summary
Abstract
Bone-marrow aspirate differential cell counts are important in the classification of haematological disorders, but manual counting is slow and subject to observer variation. This study develops a machine-learning system for detecting and classifying cells in digitised bone-marrow aspirate smears.
Pipeline
More than 10,000 cells were manually annotated using a web-based digital-pathology system. The model uses a two-stage design: first locating cells in whole-slide imagery, then classifying the detected cells into the categories used for clinical differential counts.
Results
In six-fold cross-validation on non-neoplastic samples, the system achieved a precision–recall AUC of 0.959 for detection and a ROC AUC of 0.982 for classification. Tests on a small set of acute myeloid leukaemia and multiple myeloma samples showed similar early performance.
Why it matters
The work demonstrates the feasibility of objective, machine-assisted bone-marrow differential counting and provides an early foundation for clinically validated automated workflows.
Use this work
Citation
Chandradevan, R., Aljudi, A. A., Drumheller, B. R., Kunananthaseelan, N., Amgad, M., Gutman, D. A., Cooper, L. A. D., & Jaye, D. L. (2020). Machine-based detection and classification for bone marrow aspirate differential counts: Initial development focusing on nonneoplastic cells. Laboratory Investigation, 100(1), 98–109.
@article{chandradevan2020machine,
title = {Machine-Based Detection and Classification for Bone Marrow Aspirate Differential Counts: Initial Development Focusing on Nonneoplastic Cells},
author = {Chandradevan, Ramraj and Aljudi, Ahmed A. and Drumheller, Bradley R. and Kunananthaseelan, Nilakshan and Amgad, Mohamed and Gutman, David A. and Cooper, Lee A. D. and Jaye, David L.},
journal = {Laboratory Investigation},
volume = {100},
number = {1},
pages = {98--109},
year = {2020},
doi = {10.1038/s41374-019-0325-7}
}