Automated bone marrow cytology using deep learning to generate a histogram of cell types

Area of Science:

• Computational pathology
• Artificial intelligence in medicine
• Hematology

Background:

• Bone marrow cytology is crucial for hematological diagnoses but is labor-intensive and prone to variability.
• Current limitations in bone marrow cytology can lead to diagnostic delays or errors.
• There is a need for advanced technologies to support bone marrow cytology analysis.

Purpose of the Study:

• To develop an automated, deep learning-based system for bone marrow cytology analysis.
• To create a quantitative representation of cytomorphological data called Histogram of Cell Types (HCT).
• To enhance the efficiency and accuracy of hematological diagnoses.

Main Methods:

• An end-to-end deep learning system was developed for analyzing digital whole slide images of bone marrow aspirates.
• The system automatically detects regions of interest and identifies/classifies all bone marrow cells.
• A Histogram of Cell Types (HCT) was generated to represent cell type probability distributions.

Main Results:

• The system demonstrated high accuracy in region detection (0.97 accuracy, 0.99 ROC AUC).
• Cell detection and classification achieved a mean average precision of 0.75 and an average F1-score of 0.78.
• The Log-average miss rate for cell analysis was 0.31.

Conclusions:

• The developed AI system shows significant potential for automating bone marrow cytology.
• The Histogram of Cell Types (HCT) can serve as a valuable tool for supporting hematological diagnoses.
• This technology advances AI-enabled computational pathology for more efficient and accurate diagnostics.