Paper grades cone beam CT report generation mostly by LLM entailment score

A new arXiv preprint describes a method for generating maxillofacial radiology reports from cone beam CT scans, evaluated with a composite metric that assigns 80% of the weight to a large language model's judgment of factual entailment and 20% to lexical overlap. The abstract notes that only the lexical portion of this objective is fully observable or directly optimizable. The work sits at the intersection of medical imaging report generation and LLM-based evaluation.