Search papers, labs, and topics across Lattice.
Universit脿 Campus Bio-Medico di Roma
4
0
6
High report quality in VLMs can mask a dangerous reliance on visual shortcuts, revealing a critical flaw in how we evaluate radiology report generation.
A new multimodal deep learning model accurately predicts cancer treatment response from incomplete patient data, even when key clinical information is missing.
Injecting retrieved anatomical priors into text-to-CT generation dramatically improves image fidelity and clinical consistency, offering a scalable path to more realistic medical image synthesis.
Radiology report generation can be both more accurate AND more interpretable: CEMRAG uses visual concepts to enhance multimodal RAG, challenging the assumed trade-off between transparency and diagnostic accuracy.
