Automatic computational classification of bone marrow cells for B cell pediatric leukemia using UMAP

Area of Science:

• Biomedicine
• Computational Biology
• Artificial Intelligence in Healthcare

Background:

• Pediatric B Acute Lymphoblastic Leukemia (B-ALL) is common, with 15-20% of children relapsing despite current treatments.
• Flow cytometry is crucial for monitoring residual disease in B-ALL, but manual gating is inefficient.
• Advanced computational tools, including machine learning and UMAP, offer potential for improved analysis of complex biological data.

Purpose of the Study:

• To develop an artificial intelligence-based algorithm for enhanced patient classification and therapy decisions in B-ALL.
• To overcome limitations of traditional manual gating in flow cytometry analysis for B-ALL.
• To improve the prediction and prognosis of B-ALL by identifying key subpopulations and regeneration patterns.

Main Methods:

• Utilized 234 samples from 75 B-ALL patients.
• Developed an artificial intelligence (AI)-based algorithm integrating Uniform Manifold Approximation and Projection (UMAP) and Machine Learning.
• Applied the algorithm for automated identification of key subpopulations and bone marrow regeneration patterns.

Main Results:

• The AI algorithm demonstrated improved patient classification and therapy decision-making capabilities.
• Successfully identified distinct bone marrow regeneration patterns in different patient cohorts.
• Showcased the potential of AI and mathematical tools to advance B-ALL analysis beyond traditional methods.

Conclusions:

• The developed AI algorithm represents an advancement over routine manual analysis for B-ALL disease progression monitoring.
• Automated identification of subpopulations and regeneration patterns can significantly improve B-ALL prognosis and prediction.
• This approach highlights the value of integrating advanced mathematical tools and AI in biomedical research for better patient outcomes.