Radiology

Radiomics-AI predicts VETC on preop imaging and links to early recurrence

Radiomics/deep-learning models using contrast-enhanced MRI showed pooled sensitivity 0.84 and specificity 0.79 for predicting CD34-confirmed VETC in HCC. AI-predicted VETC positivity was associated with higher risk of early recurrence (pooled HR 2.34). Evidence came from 15 retrospective studies (729 internal and 613 external validation patients) with heterogeneity and low-to-moderate GRADE certainty. These tools may aid preoperative risk stratification but require prospective validation before changing surgical management. 1

Modified Zelen design broadly acceptable to THA patients after debriefing

Among 235 CAPS-THA respondents, 76% agreed the delayed-consent design was justified and 91% judged the study ethical. Trust in medical research was unchanged or increased for 94% after disclosure. Seventy-one percent reported willingness to join a similar future study. Findings support responsible use of modified Zelen designs with robust debriefing and transparency during follow-up. 2

Generative AI clinical support was safe but did not lower short-term treatment failure

In a pragmatic cluster RCT of 9,691 primary care patients, treatment failure occurred in 2.2% with LLM assistance versus 2.0% without (adjusted OR 0.77, 95% CI 0.55–1.08, P=0.13). No serious adverse events were attributed to the intervention and independent review found no safety signal. The trial found no statistically significant reduction in 14-day treatment failure; any benefit is likely modest. LLM assistance appears safe in this setting but is not ready to deliver clear clinical benefit without further study. 3