Related Experiment Video
Updated: Oct 12, 2025
05:52
Author Spotlight: Efficient Retinal Ganglion Cell Counting in Mouse Models of Glaucoma for Treatment Evaluation
Published on: October 4, 2024
1.3K
Automatic Detection and Counting of Blood Cells in Smear Images Using RetinaNet
Grzegorz Drałus1, Damian Mazur1, Anna Czmil1
1Department of Electrical and Computer Engineering Fundamentals, Rzeszow University of Technology, 35-959 Rzeszow, Poland.
Entropy (Basel, Switzerland)
|November 27, 2021
Summary
This study introduces an automated deep learning method for accurate blood cell counting, replacing time-consuming manual techniques. The RetinaNet model effectively identifies and quantifies red blood cells, white blood cells, and platelets in microscopic images.
Area of Science:
- Medical Imaging
- Computational Biology
- Artificial Intelligence
Background:
- Complete blood count (CBC) is crucial for disease diagnosis.
- Conventional blood cell counting methods (manual, hemocytometer) are laborious and time-intensive.
- There is a need for automated, accurate blood cell analysis.
Purpose of the Study:
- To develop an automatic software-based method for accurate blood cell counting.
- To utilize the RetinaNet deep learning network for object recognition and classification in microscopic images.
- To evaluate the performance and generalizability of the proposed automated counting system.
Main Methods:
- Implementation of the RetinaNet deep learning network for microscopic image analysis.
- Training the network to automatically recognize and count red blood cells, white blood cells, and platelets.
- Assessment of detection and counting quality using accuracy, sensitivity, precision, and F1-score.
- Analysis of confidence thresholds and learning epochs' impact on results.
- Comparison with existing cell counting methodologies.
Main Results:
- The trained RetinaNet model demonstrated generalized capabilities on smear images.
- Performance metrics (accuracy, sensitivity, precision, F1-score) validated the model's effectiveness.
- The study analyzed the influence of confidence thresholds and learning epochs on recognition and counting outcomes.
- The proposed method showed competitive performance compared to other authors' approaches.
- Object detection and labeling were identified as beneficial additions to cell counting.
Conclusions:
- The proposed automated software-based method offers an accurate alternative to conventional blood cell counting.
- Deep learning, specifically RetinaNet, is effective for automated recognition and quantification of blood cells.
- The system's generalizability and performance metrics support its clinical potential.
- Object detection and labeling enhance the utility of automated cell counting systems.
