Deep Learning Enables Automated Detection of Mature B Cell Neoplasms by Flow Cytometry

Area of Science:

• Computational Biology
• Hematology
• Artificial Intelligence

Background:

• Multiparameter flow cytometry is crucial for diagnosing mature B-cell lymphomas.
• Current manual analysis is time-consuming and prone to variability.

Purpose of the Study:

• To develop an automated deep learning system for B-cell neoplasm detection.
• To improve diagnostic accuracy and efficiency in flow cytometry analysis.

Main Methods:

• A three-stage deep learning architecture was employed.
• The system utilizes light-chain, cell-level, and sample-level modules.
• Evaluated on 3,070 clinical specimens including peripheral blood, bone marrow, and tissue samples.

Main Results:

• Achieved 97.5% case-level accuracy, 95.6% sensitivity, and 98.9% specificity.
• Predicted abnormal cell counts strongly correlated with expert manual counts (R²=0.962).
• Features supported lymphoma subtype prediction with an AUROC of 0.925.

Conclusions:

• Automated flow cytometry interpretation can achieve high diagnostic accuracy.
• The system provides transparent, verifiable outputs suitable for clinical integration.
• Deep learning enhances diagnostic capabilities for mature B-cell lymphomas.