A deep learning method for counting white blood cells in bone marrow images.

Area of Science:

• Hematology
• Medical Imaging
• Artificial Intelligence

Background:

• Manual differentiation and counting of white blood cells (WBC) in bone marrow smears are crucial for diagnosing various conditions like infection, anemia, and leukemia.
• Manual methods are time-consuming, fatiguing, and dependent on operator expertise, impacting accuracy.

Purpose of the Study:

• To develop and evaluate an automated deep learning system for accurate and efficient classification and counting of WBC in bone marrow microscopic images.
• To address the limitations of manual cell counting, including time consumption and operator variability.

Main Methods:

• A deep learning approach utilizing Faster R-CNN and Feature Pyramid Network was employed for automatic cell counting.
• The system was designed to handle variations in illumination and maintain color component stability.
• The model was trained and tested on a dataset from The Second Affiliated Hospital of Zhejiang University.

Main Results:

• The automated system achieved a high overall correct recognition rate of up to 98.8% on a dataset of 609 bone marrow images.
• The deep learning model demonstrated effectiveness in classifying and counting different types of white blood cells.
• Performance was comparable to existing state-of-the-art systems.

Conclusions:

• The proposed deep learning system offers a highly accurate and efficient solution for white blood cell classification and counting.
• The system's user interface facilitates easy operation by pathologists.
• This technology has the potential to significantly improve diagnostic workflows in hematology.