Validation of Artificial Intelligence (AI)-Assisted Flow Cytometry Analysis for Immunological Disorders

Area of Science:

• Clinical Immunology
• Computational Biology
• Hematology

Background:

• Flow cytometry is crucial for diagnosing hematologic and immunologic disorders.
• Manual analysis of flow cytometry data is time-consuming and subject to variability.
• Advancements in artificial intelligence (AI) offer potential solutions for automated analysis.

Purpose of the Study:

• To develop and validate an AI-assisted flow cytometry workflow for diagnosing primary immunodeficiency and related immunological disorders.
• To assess the accuracy and efficiency of the AI system compared to manual analysis.
• To reduce the time required for flow cytometry data interpretation.

Main Methods:

• Utilized a 3-tube, 10-color flow panel with 21 antibodies across 379 clinical cases.
• Developed and implemented an automated AI software (DeepFlow™, version 2.1.1) employing a multidimensional density-phenotype coupling algorithm.
• Validated AI performance against manual analysis by hematopathologists, using lymphocyte subset percentages as the gold standard.

Main Results:

• The AI workflow reduced analysis time to under 5 minutes per case.
• The AI model accurately classified and enumerated T, B, and NK cells, including critical subsets like CD4+, CD8+, double-negative T cells, and switched/non-switched B cells.
• A strong correlation (r > 0.9) was observed between AI-derived and manually determined lymphocyte subset percentages.

Conclusions:

• AI-assisted flow cytometry provides an accurate and efficient method for diagnosing immunological disorders.
• The developed AI workflow offers a transformative approach, significantly reducing analysis time in a clinical setting.
• This technology enhances diagnostic capabilities for hematopathologists, improving patient care.