NOVA: A Benchmark for Anomaly Localization and Clinical Reasoning in Brain MRI
Cosmin I. Bercea, Jun Li, Philipp Raffler, Evamaria O. Riedel, Lena Schmitzer, Angela Kurz, Felix Bitzer, Paula Roßmüller, Julian Canisius, Mirjam L. Beyrle, Che Liu, Wenjia Bai, Bernhard Kainz, Julia A. Schnabel, Benedikt Wiestler
2025-05-28
Summary
This paper talks about NOVA, a new test that checks how well AI systems can find unusual problems in brain MRI scans and explain their reasoning, especially when those problems are rare or not seen before.
What's the problem?
The problem is that most AI models aren't very good at spotting rare or unexpected brain issues in MRI images, and they often struggle to explain what they find, which is a big deal in medicine where doctors need to trust and understand AI decisions.
What's the solution?
The researchers created the NOVA benchmark, which includes lots of examples of rare brain conditions in MRI scans, to see how well AI models can recognize and describe these unusual cases using both images and language.
Why it matters?
This matters because it helps make AI better at supporting doctors in diagnosing tough or rare medical problems, making healthcare safer and more reliable for patients.
Abstract
NOVA is a benchmark for evaluating vision-language models on rare, clinically relevant MRI pathologies, challenging their out-of-distribution and open-world recognition capabilities.