TUM team frames PET scans as visual Q&A to predict 24-month survival in PSMA prostate cancer therapy
maier_ak · x · 2026-10-20
A TUM and Bern University Hospital team published a study in the Journal of Nuclear Medicine (first author Mohamed Kassar, senior author Kuangyu Shi) framing survival prognosis after PSMA radioligand therapy as visual question answering: baseline and interim PET/SPECT scans, PSA values, and a clinical query go in, a 24-month survival answer comes out. A 3D image encoder processes the scans, RadBERT encodes the query, fused via cross-modal attention — mirroring how physicians compare baseline and interim reads.
- 81.8% accuracy on 110 metastatic prostate cancer patients, but the 95% CI spans 0.52–0.95
- External validation: 71.9% after few-shot fine-tuning, AUC 0.75 — best among compared models
- A query-free fusion model performs comparably, so the question framing is not yet shown to matter; the authors rightly call it a proof of concept
Related event: Multimodal VQA Model Predicts Prostate Cancer Survival After Therapy(2 posts)→
More from AGI Musings
- UK government advisers warn Britain risks becoming a 'high-value AI customer' — Polymarket · 2026-10-20
- Mathematician says AI proved a combinatorics conjecture, now wonders where to share it — tak3sh8 · 2026-10-20
- Altman reportedly outlined three superintelligence endgames to OpenAI staff, including one lab locking out all rivals — kimmonismus · 2026-10-20
- School trains you to fear mistakes, not to survive being ignored — birchlse · 2026-10-20
- Jason Wei: A Year Obsessed With RL Taught Me an On-Policy Lesson About Life — paigeinsf · 2026-10-20
- Anna Ciaunica argues consciousness is a journey, not a thing — and why AI cannot have it — AnnaCiaunica · 2026-10-20