Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
KAIST
2
0
4
2
Despite possessing relevant medical knowledge, current MLLMs fail spectacularly at multi-step ECG reasoning, achieving a mere 6% completion rate due to their inability to link clinical criteria to visual ECG evidence.
Clinically-grounded diagnostic agents that reason about chest X-rays outperform large vision-language models in faithfulness and verifiability, without requiring costly retraining for new tasks.