Predicting Breast Cancer Receptor Status and Molecular Subtype Using MRI Images

Context & Background

Breast cancer therapy is highly dependent on identifying receptor statuses (ER, PR, HER2). Currently, receptor status is determined via invasive tissue biopsies, which are expensive, painful, and carry risks.

Problems to be Addressed

Biopsies are invasive and might miss spatial heterogeneity in tumors. Non-invasive radiomics models are needed to predict molecular subtypes directly from magnetic resonance imaging (MRI).

Aims and Objectives

1. Develop deep learning models to predict breast cancer receptor status from MRI scan sequences.
2. Build non-invasive radiomics feature classification pipelines.
3. Validate predictions using patient cohorts from Max Healthcare.

Methodology

The team uses 3D CNN architectures and ResNet variants to extract spatial features from multi-parametric breast MRI scans. The models are validated against ground-truth histopathology reports to ensure high specificity and sensitivity.

Expected Outcomes

A non-invasive diagnostic classifier scoring breast cancer subtypes, radiomics APIs, and clinical validation studies.