Related Experiment Video
Updated: Jun 7, 2025
09:31
Automated Quantification of Hematopoietic Cell – Stromal Cell Interactions in Histological Images of Undecalcified Bone
Published on: April 8, 2015
11.5K
A machine-learning-based algorithm for bone marrow cell differential counting
Ta-Chuan Yu1, Cheng-Kun Yang2, Wei-Han Hsu2
1Department of Internal Medicine, National Taiwan University Hospital Yunlin Branch, No. 579, Sec. 2, Yunlin Rd., Douliu City, Yunlin County 640203, Taiwan.
International Journal of Medical Informatics
|November 16, 2024
Summary
This study introduces an AI algorithm for automated bone marrow differential counting, achieving high accuracy in identifying and classifying various cell types. The developed AI shows potential for clinical application in diagnosing hematological diseases.
Area of Science:
- Hematology
- Medical Imaging
- Artificial Intelligence
Background:
- Bone marrow differential counting is vital for diagnosing hematological diseases.
- Current manual methods are time-consuming and subjective.
- A need exists for an automated, clinically applicable solution for differential counting.
Purpose of the Study:
- To develop and validate an artificial intelligence (AI)-based algorithm for automated identification and classification of nucleated cells in bone marrow smears.
- To assess the clinical applicability of the AI algorithm for bone marrow differential counting.
Main Methods:
- A Mask R-CNN-based AI model was trained on a large dataset of expert-annotated bone marrow cell images.
- Images were stained with Liu's stain or Wright-Giemsa stain, with expert consensus ensuring consistent classification criteria.
- The AI algorithm's performance was evaluated on multinational clinical datasets for cell identification and differential counting ratios.
Main Results:
- The AI model achieved an accuracy of 0.94 on a testing dataset and 0.881 on a multinational real-world dataset.
- High precision was observed in classifying specific cell types, including blasts (0.927) and neutrophils (0.955).
- Strong correlations (ρ > 0.8) were found between AI and manual differential counting percentages for most cell types.
Conclusions:
- An AI algorithm for bone marrow differential counting has been successfully developed and clinically validated.
- The algorithm can simultaneously locate and classify bone marrow cells.
- This AI tool holds significant potential for automating bone marrow differential counting in clinical practice.
