Materials for two hands-on student challenges (Summer School in AI for Cancer Research), built on the same clinical question: at the response evaluation of the first TACE cycle in hepatocellular carcinoma, should the patient receive further TACE or switch to systemic therapy?
Everything runs on pre-extracted CT radiomic features + clinical variables from the open WAW-TACE dataset — a few MB of tabular data, no GPU, no imaging expertise needed. Each challenge lives on its own branch; your facilitator will tell you which one you are doing.
🔬 Challenge 1 — Model development & validation (branch dev)
Can we predict who will respond to TACE? You are the research team: build and interpret models for two targets — progression classification against a clinical-only baseline, and an independent Cox model of overall survival — then decide what your model can honestly claim.
🔍 Challenge 2 — The FUTURE-AI audit (branch audit)
Would you trust this model? You are the clinical AI review board: a frozen, pre-trained model arrives with a model card and a strong AUC. Audit it along the six FUTURE-AI principles and deliver a verdict — deploy, fix, or reject.
- Data: WAW-TACE dataset (CC-BY-4.0) — Bartnik K, et al., Medical University of Warsaw. The CSVs on both branches are redistributed derivatives (clinical table and pre-extracted radiomics features) under the same license.
- External cohort: HCC-TACE-Seg — Moawad AW, et al., Sci Data 2023;10:33, The Cancer Imaging Archive / MD Anderson.
- Radiomics pipeline adapted from CCIG-Champalimaud/waw-tace-radiomics.
- Audit framework: Lekadir K, et al. FUTURE-AI: international consensus guideline for trustworthy and deployable artificial intelligence in healthcare. BMJ 2025;388:e081554.
Educational material. All models involved are teaching artifacts — they are not medical devices and must not inform clinical decisions.