Radiomics 9
Radiomics: Foundation-Model Radiomics for Cancer Prognostics (GenAI + Agentic RAG)
Build AI-driven radiomics pipelines linking imaging features to clinical and genetic outcomes.
Mentor Details:
Prof. Ilan Tsarfaty
Mentor Details:
Requirments:
Combine classical radiomics with foundation-model embeddings for improved prognostic performance.
Problem Statement
Imaging-derived biomarkers are underutilized for prognostic modeling.
Project Objectives
Combine classical radiomics with foundation-model embeddings for improved prognostic performance.
Technical Scope
Build an AI-driven radiomics pipeline that extracts quantitative imaging features from clinical/preclinical scans and links them to genetic and clinical outcomes. You will work with curated imaging datasets, apply foundation-model feature extraction, and implement interpretable evaluations for prognostic tasks.
Required Knowledge and Prerequisites
Core Requirements:
Python, ML, statistics.
Recommended Background:
Medical imaging, model interpretability.
Project Difficulty and Expected Level
Overall Difficulty: High
This project is well-suited for:
Teams of 2–4 students
Expected Outcomes
Practical competency in medical imaging AI: preprocessing, feature engineering at scale, model validation, and bias/robustness considerations.
Experience combining classic radiomics with modern foundation model embeddings for improved generalization.
Hands-on practice building usable tooling (API/GUI) rather than “one-off notebooks.