Radiologists play a crucial role in translating medical images into actionable reports. However, the field faces staffing shortages and increasing workloads. While automated approaches using vision-language models (VLMs) show promise as assistants, they require exceptionally high accuracy. Most current VLMs in radiology rely solely on supervised fine-tuning. Meanwhile, additional preference fine-tuning in the post-training pipeline has become standard practice in the general domain. The challenge in radiology lies in the prohibitive cost of obtaining radiologist feedback at scale. To address this challenge, we propose an automated pipeline for preference feedback, focusing on chest X-ray radiology report generation (RRG). Specifically, our method leverages publicly available datasets containing pairs of images and radiologist-written reference reports with reference-based metrics, or Judges, eliminating the need for *additional radiologist feedback*. We investigate reward overoptimization via length exploitation in this setting and introduce a length-controlled version of the GREEN score. Our best-performing setup achieves state-of-the-art CheXbert scores on the MIMIC-CXR dataset for the RRG task while on average maintaining robust performance across six additional image perception and reasoning tasks.

Evaluating radiology reports is a challenging problem as factual correctness is extremely important due to its medical nature. Existing automatic evaluation metrics either suffer from failing to consider factual correctness (e.g., BLEU and ROUGE) or are limited in their interpretability (e.g., F1CheXpert and F1RadGraph). In this paper, we introduce GREEN (Generative Radiology Report Evaluation and Error Notation), a radiology report generation metric that leverages the natural language understanding of language models to identify and explain clinically significant errors in candidate reports, both quantitatively and qualitatively. Compared to current metrics, GREEN offers: 1) a score aligned with expert preferences, 2) human interpretable explanations of clinically significant errors, enabling feedback loops with end-users, and 3) a lightweight open-source method that reaches the performance of commercial counterparts. We validate our GREEN metric by comparing it to GPT-4, as well as to error counts of 6 experts and preferences of 2 experts. Our method demonstrates not only higher correlation with expert error counts, but simultaneously higher alignment with expert preferences when compared to previous approaches.

Motivated by the scarcity of high-quality labeled biomedical text, as well as the success of data programming, we introduce KRISS-Search. By leveraging the Unified Medical Language Systems (UMLS) ontology, KRISS-Search addresses an interactive few-shot span recommendation task that we propose. We first introduce unsupervised KRISS-Search and show that our method outperforms existing methods in identifying spans that are semantically similar to a given span of interest, with >50% AUPRC improvement relative to PubMedBERT. We then introduce supervised KRISS-Search, which leverages human interaction to improve the notion of similarity used by unsupervised KRISS-Search. Through simulated human feedback, we demonstrate an enhanced F1 score of 0.68 in classifying spans as semantically similar or different in the low-label setting, outperforming PubMedBERT by 2 F1 points. Finally, supervised KRISS-Search demonstrates competitive or superior performance compared to PubMedBERT in few-shot biomedical named entity recognition (NER) across five benchmark datasets, with an average improvement of 5.6 F1 points. We envision KRISS-Search increasing the efficiency of programmatic data labeling and also providing broader utility as an interactive biomedical search engine.