Comparison of three machine learning algorithms for classification of B-cell neoplasms using clinical flow cytometry...

Area of Science:

• Clinical diagnostics
• Bioinformatics
• Machine learning in healthcare

Background:

• Multiparameter flow cytometry is crucial for clinical disease diagnosis but is labor-intensive.
• Artificial intelligence (AI) offers potential to automate and improve the interpretation of flow cytometry data.
• Previous studies have explored machine learning for flow cytometry data classification.

Purpose of the Study:

• To evaluate three machine learning methods for classifying flow cytometry data.
• To apply these methods to a B-cell neoplasm dataset for disease subtype prediction.
• To compare algorithm performance based on accuracy and computational time.

Main Methods:

• Examined three established machine learning algorithms for flow cytometry data classification.
• Applied algorithms to ungated flow cytometry data from a B-cell neoplasm cohort.
• Compared algorithm architectures, multiclass classification accuracies, and computation times.

Main Results:

• Two methods, flowCat and EnsembleCNN, achieved comparable high accuracies in classifying B-cell neoplasms.
• Both flowCat and EnsembleCNN demonstrated relatively fast computational performance.
• EnsembleCNN exhibited a speed advantage, especially when incorporating additional training data.

Conclusions:

• AI algorithms, specifically flowCat and EnsembleCNN, show promise in assisting the clinical interpretation of flow cytometry data.
• These AI methods can accurately predict disease subtypes from B-cell neoplasm data.
• EnsembleCNN presents an efficient and scalable solution for flow cytometry data analysis.