Automated Immunophenotyping Assessment for Diagnosing Childhood Acute Leukemia using Set-Transformers

Area of Science:

• Hematology
• Oncology
• Bioinformatics

Background:

• Childhood acute leukemia is a common pediatric hematologic malignancy.
• Immunophenotyping via Multiparameter Flow Cytometry (FCM) is crucial for diagnosis.
• Current FCM methods are manual, time-consuming, and subjective.

Purpose of the Study:

• To develop an automated, accurate, and objective tool for immunophenotyping in pediatric acute leukemia.
• To introduce the FCM-Former, a self-attention based machine learning model for FCM data analysis.

Main Methods:

• The FCM-Former model was developed using supervised learning with flow cytometric data.
• The model employs a self-attention mechanism for enhanced data processing.
• The tool was trained and validated on a dataset of 960 pediatric acute leukemia cases.

Main Results:

• The FCM-Former achieved a diagnostic accuracy of 96.5% in classifying leukemia subtypes (B-ALL, T-ALL, AML).
• The model demonstrated high performance in assigning lineage to each sample.
• This represents the first automated immunophenotyping assessment tool utilizing FCM data for pediatric acute leukemia.

Conclusions:

• The FCM-Former offers a significant advancement in the automated diagnosis of pediatric acute leukemia.
• This AI-driven tool can improve the efficiency and objectivity of immunophenotyping.
• FCM-Former has the potential to streamline diagnostic workflows and improve patient outcomes.