Augmented Human Intelligence and Automated Diagnosis in Flow Cytometry for Hematologic Malignancies

Area of Science:

• Hematology
• Computational Biology
• Medical Diagnostics

Background:

• Clinical flow cytometry is resource-intensive, requiring expert review and significant time.
• The manual analysis of flow cytometry data, especially for B-cell malignancies, presents challenges in efficiency and cost.
• Automation in flow cytometry analysis is crucial for improving diagnostic workflows.

Purpose of the Study:

• To develop and evaluate an automated pipeline for diagnosing B-cell malignancies using flow cytometry data.
• To assess the performance of the automated system against current clinical diagnostic standards.
• To explore the potential of automation for increasing efficiency and reducing the burden on laboratory personnel.

Main Methods:

• Utilized 3,417 peripheral blood flow cytometry cases analyzed over six months.
• Applied Uniform Manifold Approximation and Projection (UMAP) for feature extraction and dimensionality reduction on raw data.
• Employed random forest classification for B-cell malignancy diagnosis, bypassing traditional gating methods.

Main Results:

• The automated classifier demonstrated over 95% accuracy in diagnosing B-cell malignancies.
• High accuracy was observed for specific conditions like chronic lymphocytic leukemia.
• Adjustable cutoffs theoretically enable 100% sensitivity with 14% specificity, potentially allowing 11% of cases for automated verification.

Conclusions:

• The developed automated pipeline can significantly improve quality control in clinical flow cytometry.
• Implementation of this system is expected to reduce diagnostic turnaround times.
• Automation holds the potential to decrease staff workloads and enhance overall laboratory efficiency.