Related Experiment Video
Updated: Nov 16, 2025
11:59
Detection of Residual Donor Erythroid Progenitor Cells after Hematopoietic Stem Cell Transplantation for Patients with Hemoglobinopathies
Published on: September 6, 2017
7.6K
Improving blood cells classification in peripheral blood smears using enhanced incremental training.
Rabiah Al-Qudah1, Ching Y Suen1
1Department of Computer Science, Concordia University, 1455 Boulevard de Maisonneuve O, Montréal, QC, Canada.
Computers in Biology and Medicine
|February 23, 2021
Summary
Automated peripheral blood smear analysis using deep learning significantly improves the classification of white blood cells and platelets. This AI approach enhances accuracy and precision, crucial for medical diagnostics and worker safety.
Area of Science:
- Medical Diagnostics
- Artificial Intelligence in Healthcare
- Computational Pathology
Background:
- Peripheral Blood Smear (PBS) analysis is a critical routine medical test.
- Automating PBS analysis offers benefits like time/cost savings, error reduction, and enhanced safety for healthcare workers, especially during pandemics.
- Current methods face challenges in accurately classifying various blood cell subtypes and abnormalities.
Purpose of the Study:
- To develop and evaluate deep learning models for automated classification of fifteen white blood cell and platelet subtypes and morphological abnormalities from synthetic blood smears.
- To improve the accuracy and precision of automated blood cell classification.
- To enhance the safety and efficiency of medical laboratory diagnostics.
Main Methods:
- Training deep neural networks on a synthetic blood smear dataset.
- Implementing a hybrid deep learning and image processing approach for platelet classification.
- Developing a novel Enhanced Incremental Training scheme and a 'training revert' procedure for white blood cell classification to handle confusable classes.
Main Results:
- The hybrid approach for platelet classification achieved an accuracy of 98.6% and macro-average precision of 97.6%.
- The novel methods for white blood cell classification improved accuracy from 61.5% to 95% and macro-average precision from 76.6% to 94.27%.
- The developed deep learning models demonstrated significant improvements in classifying various blood cell subtypes and abnormalities.
Conclusions:
- Deep learning models, particularly with novel training strategies, show high potential for accurate and efficient automated peripheral blood smear analysis.
- The proposed methods offer a robust solution for classifying complex blood cell subtypes and abnormalities, outperforming previous benchmarks.
- Automated PBS analysis using AI can revolutionize medical diagnostics, improving accuracy and protecting healthcare professionals.
