Automated Flow Cytometric MRD Assessment in Childhood Acute B- Lymphoblastic Leukemia Using Supervised Machine...

Area of Science:

• Hematology
• Computational Biology
• Medical Technology

Background:

• Minimal residual disease (MRD) detection via multiparameter flow cytometry (FCM) is crucial for B-cell acute lymphoblastic leukemia (B-ALL) prognosis.
• Current FCM-MRD quantification relies heavily on operator expertise, introducing subjectivity and variability.
• A need exists for objective, automated tools to standardize FCM-MRD analysis.

Purpose of the Study:

• To develop and validate a supervised machine learning approach for automated FCM-MRD quantification in pediatric B-ALL.
• To assess the performance of the developed method against expert assessments and other automated techniques.
• To evaluate the cross-laboratory and cross-system generalizability of the automated approach.

Main Methods:

• A supervised machine learning model combining multiple Gaussian Mixture Models (GMMs) was developed for FCM-MRD quantification.
• The model uses GMMs to represent sample data and interpolates stored samples to analyze new data.
• The approach was trained and tested on 337 pediatric B-ALL bone marrow samples from three laboratories.

Main Results:

• The GMM-combination approach demonstrated superior precision and F1-scores compared to support vector machines, deep neural networks, and single GMM methods.
• Automated MRD quantification achieved high correlation with expert assessments, with F1-scores >0.5 in over 95% of samples.
• Cross-system performance remained robust, with median F1-scores above 0.85 even when data originated from different flow cytometry systems.

Conclusions:

• The proposed automated GMM-combination approach offers an objective and standardized method for FCM-MRD assessment in B-ALL.
• This tool has the potential to improve the reliability and consistency of MRD quantification across diverse laboratory settings.
• Standardized MRD assessment can enhance prognostic accuracy and guide treatment decisions in B-ALL patients.