Automated CLL cell population detection using a weakly supervised approach and CLL MRD flow cytometry data.

Area of Science:

• Oncology
• Bioinformatics
• Machine Learning

Background:

• Minimal/measurable residual disease (MRD) detection is crucial for cancer treatment monitoring.
• Multiparameter flow cytometry is a common method for MRD assessment.
• Current methods may require extensive annotation or complex analysis.

Purpose of the Study:

• To propose a novel machine learning approach for binary prediction of MRD status using flow cytometry data.
• To develop a weakly supervised method requiring only neoplastic cell percentage for training.
• To evaluate the method's accuracy and applicability in a chronic lymphocytic leukemia cohort.

Main Methods:

• Projection of cells into a low-dimensional embedding space.
• Clustering cells by similarity within the embedded space.
• Utilizing cluster-wise cell proportions for prediction and regression.

Main Results:

• High accuracy achieved in predicting MRD status for chronic lymphocytic leukemia patients.
• Demonstrated the effectiveness of the weakly supervised learning approach.
• Successfully applied dimensionality reduction and clustering for MRD prediction.

Conclusions:

• The proposed machine learning method offers an accurate and efficient approach for MRD detection.
• Weakly supervised learning reduces annotation burden in flow cytometry data analysis.
• This method holds promise for routine clinical application in cancer monitoring.