Experimental evaluation of deep learning method in reticulocyte enumeration in peripheral blood
Geng Wang1, Tianci Zhao1, Zhejun Fang2
1Department of Clinical Laboratory, Peking Union Medical College Hospital, Beijing, China.
International Journal of Laboratory Hematology
|May 20, 2021
Summary
A deep learning model accurately counts reticulocytes (RET), immature red blood cells, improving upon manual methods. This AI approach offers a faster, more objective tool for clinical diagnostics in anemia.
Area of Science:
- Hematology
- Medical Imaging
- Artificial Intelligence
Background:
- Reticulocyte (RET) enumeration is crucial for diagnosing and monitoring anemia.
- Current methods, flow cytometry and manual microscopy, have limitations in accuracy and efficiency.
- There is a need for an objective, precise, and rapid method for RET counting.
Purpose of the Study:
- To develop and evaluate a deep learning model for automated reticulocyte enumeration.
- To assess the accuracy and speed of the deep learning model compared to existing methods.
Main Methods:
- A Faster R-CNN deep neural network was trained on 784 labeled microscopic images of red blood cells.
- The dataset comprised 40 whole blood samples.
- The model was evaluated for recall, precision, and analysis time.
Main Results:
- The deep learning model achieved recall and precision rates exceeding 97%.
- The average analysis time per image was 0.21 seconds.
- High accuracy and speed were demonstrated.
Conclusions:
- Deep learning offers a highly accurate and rapid method for reticulocyte enumeration.
- This AI-driven approach has the potential to serve as a valuable computer-aid for cytological examiners.
- The method addresses the limitations of current manual and flow cytometric techniques.
