Related Experiment Video
Updated: Jul 13, 2025
09:31
Automated Quantification of Hematopoietic Cell – Stromal Cell Interactions in Histological Images of Undecalcified Bone
Published on: April 8, 2015
11.6K
Virtual Staining, Segmentation, and Classification of Blood Smears for Label-Free Hematology Analysis
Nischita Kaza1, Ashkan Ojaghi2, Francisco E Robles2
1School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, Georgia, USA.
BME Frontiers
|October 18, 2023
Summary
A novel deep-ultraviolet microscopy framework offers automated, label-free hematological analysis. This cost-effective method simplifies blood cell counting and classification, paving the way for point-of-care diagnostics.
Area of Science:
- Biomedical Engineering
- Microscopy
- Artificial Intelligence
Background:
- Hematological analysis is crucial for disease diagnosis but relies on complex, costly, and time-consuming conventional methods.
- Label-free techniques, such as deep-ultraviolet microscopy, offer a simpler workflow by eliminating staining and preprocessing.
- Existing methods require specialized equipment and trained personnel, limiting accessibility and speed.
Purpose of the Study:
- To develop a fully automated hematological analysis framework using label-free deep-ultraviolet microscopy.
- To create a deep learning pipeline for virtual staining, segmentation, classification, and counting of white blood cells (WBCs).
- To establish a fast, cost-effective alternative to conventional hematology analyzers.
Main Methods:
- Utilized single-channel, label-free deep-ultraviolet microscopy of peripheral blood smears.
- Developed independent deep neural networks for virtual staining and cellular/nuclear segmentation.
- Trained a classifier on segmented images to perform a quantitative five-part white blood cell differential.
Main Results:
- Virtual staining accurately mimicked conventional Giemsa staining results.
- High accuracy was achieved in cellular and nuclear segmentation of blood cells.
- The classifier successfully performed a quantitative five-part WBC differential on unseen data.
Conclusions:
- The proposed automated framework simplifies and enhances complete blood count and blood smear analysis.
- This technology has the potential to enable the development of simple, fast, and low-cost point-of-care hematology analyzers.
- Label-free deep-UV microscopy combined with deep learning offers a promising avenue for advanced hematological diagnostics.
