Design and Development of Interpretable AI Methods for Flow Cytometric Immunophenotyping Data from Blood Cancers

Context & Background

Blood cancers are highly prevalent in India. Multiparametric Flow Cytometry (FCM) is the gold standard for diagnosing and monitoring leukemia and lymphoma. However, FCM generates data for millions of cells per sample, which is currently analyzed manually by pathologists. This process is time-consuming, subjective, and prone to errors.

Problems to be Addressed

FCM data gating (identifying cell subtypes in high-dimensional space) is subjective, leading to inconsistent clinical diagnoses. Existing analysis softwares lack explainability, making doctors hesitant to trust automated classifications.

Aims and Objectives

1. Formulate unsupervised clustering and gating models for FCM data.
2. Build interpretability tools to explain B-cell Acute Lymphoblastic Leukemia (B-ALL) classification.
3. Design a diagnostic dashboard for clinical testing.

Methodology

In collaboration with AIIMS, the project collects bone marrow data from leukemia patients. Machine learning models perform cell population clustering across 10-20 dimensions. The system translates these high-dimensional cell clusters into 2D visualizations, giving clinicians interactive tools to refine gating thresholds.

Expected Outcomes

A software tool for automated flow cytometry interpretation, validation with clinical cohorts at AIIMS, and patents for diagnostic classification.