FedPref lets institutions with unequal, unpooled data benefit from collaboration without ever sharing reports or annotations, and frozen public language models propose alternative JSON extractions, local annotations rank them, and sites collaboratively train compact Qwen3-8B adapters while sharing only model updates.
Abstract
Radiology reports describe findings and locations in free text, but downstream search and analysis require these relations in a fixed schema. Learning this extraction requires labels that are unevenly distributed across institutions: smaller hospitals have less local evidence, and pooling data may be infeasible. We introduce FedPref: frozen public language models propose alternative JSON extractions, local annotations rank them, and sites collaboratively train compact Qwen3-8B adapters while sharing only model updates. A heterogeneous teacher pool provides cross-model contrast when repeated single-model samples collapse. On development data from six simulated hospitals with unequal data volume and disease prevalence, FedPref improves client-mean F1 by 2.49 points and worst-site F1 by 9.10 points compared with training each site in isolation, with the largest gains at the sites holding the least data. Central training on the pooled preference-pair union is 2.66 points higher on client-mean F1. On a locked, 400-report manually validated gold test set, FedPref reaches 68.68 F1 and pooled training 71.67, preserving that same ordering. FedPref thus lets institutions with unequal, unpooled data benefit from collaboration without ever sharing reports or annotations.
Public chest-radiograph (CXR) datasets are typically released with small, fixed label schemas such as CheXpert-14. However, the underlying free-text reports describe far more findings -- and which findings matter depends on the task, site, and reader. We release a pipeline that converts free-text reports into multi-label matrices and then reconfigures the label schema through dictionary edits rather than new inference passes, i.e., without relabeling the corpus. After this one-time pass, reconfiguring MIMIC-CXR (223K reports) from cached annotations takes 196 seconds with no API cost, compared to \$6.6K for an equivalent relabeling pass with Claude Opus 4.7. Using a 58-label taxonomy, we show that 43\% of CXR studies contain at least one finding outside CheXpert-14. Image probes trained on these labels match CheXpert-14 probes on shared targets while also reaching 0.78 AUROC on expert-reviewed long-tail labels that CheXpert-14 cannot represent. These results suggest a different unit of work for radiology labeling: once reports are structured, the label schema becomes a configuration to edit, not a corpus to relabel.
Jean-Benoit Delbrouck, Dave Van Veen, A. Pattnaik et al.· 0 citations
Background Radiology impressions guide clinical care. Large Language Models (LLMs)-drafted impressions can drift into generic, off-style text. Retrieval-augmented generation (RAG) enables context-aware few-shot prompting during inference. Methods This retrospective IRB-approved study included 11,998 CT pulmonary angiography (CTPA) reports. We built a retrieval bank from 11,399 reports and reserved 599 reports for testing. GPT-4o and LLaMA 3.1-70B generated impressions from the “findings” section using three setups: zero-shot, fixed random few-shot, and dynamic retrieval-selected few-shot (top-k semantic matches; k = 3/5/10). We ran temperatures 0, 0.7, 1. We scored outputs against the original impressions with ROUGE and BERTScore F1, report mean scores with 95% confidence intervals, and tested for statistical significance using Wilcoxon signed-rank test. Results Dynamic retrieval-based few-shot prompting outperformed zero-shot and fixed few-shot prompting across all configurations (all p < 0.05). The highest scores were observed at temperature 0 and k = 10. ROUGE-1 F1 increased to 0.44–0.47 for GPT-4o and 0.37–0.50 for LLaMA, versus 0.35–0.37 and 0.25–0.37, respectively, in zero-shot prompting. Lower temperature and larger k were associated with higher similarity scores. Conclusions Dynamic, case-matched retrieval improved alignment of LLM-generated CTPA impressions with reference impressions on automated text-similarity metrics. Scores remained moderate, and radiologists’ verification is still required before clinical deployment.
Vera Sorin, Jeremy D. Collins, Lewis Hahn et al.· PLoS ONE· 0 citations
Radiology reports contain essential clinical information but often remain in unstructured, free-text formats. Notably, multiple imaging examinations performed simultaneously (such as CT head, facial bones, and cervical spine in trauma cases) may be bundled into a single report that consolidates findings from all studies into one free-text document, written jointly. Because individual sentences may reference ambiguous or overlapping anatomy (e.g., “there is a fracture”), sentence-level anatomic classification—filtering a report to retain only findings relevant to a specific anatomical region—is essential for downstream tasks such as structured label extraction and for creating clean, bijective training data for radiology report generation models. While formatting differs across reports, the clinical language remains precise. Using that fact, we develop context-aware classical models with feature engineering that surpass trained neural networks and pre-trained language models. We show that the learned model weights generalize effectively to MIMIC-IV radiology reports and that our approach achieves near-optimal performance with only a small amount of labeled training data. Together, these results make our approach practical and reproducible for new settings.
Zakk Heile, P. Manjunath, Brian Lerner et al.· Proceedings of machine learn...· 1 citation
Medical guidelines encode rich, evidence-based decision logic, yet the specific decision artifact a clinician needs is hard to locate within a guideline, let alone across guidelines covering plausible diseases and treatments. While guideline passages have supported end-to-end question answering, flowcharts remain largely underused in decision support despite their ability to encode actionable clinical pathways. We therefore introduce Case2Flow, a task designed to retrieve the most relevant guideline flowchart for a given patient case from a collection of guideline documents. To support it, we construct FlowAtlas, a curated corpus of 202 flowcharts extracted from 2,080 medical guidelines, together with a pipeline that synthesises 1,911 aligned case-flowchart pairs. Our evaluation of multimodal retrieval methods reveals systematic failure modes, including overreliance on keywords and spurious token-patch matches induced by uninformative background regions in flowcharts. Motivated by this, we propose CRISP, a training-free scoring method that sharpens late-interaction retrieval by suppressing uninformative patches, discounting ambiguous token matches, and incorporating bidirectional query-image alignment. CRISP improves Recall@1 by up to 18.71 percentage points, while a blinded physician assessment on published case narratives provides preliminary feasibility evidence beyond synthetic queries.
Jiale Wei, Yufan Chen, Alexander Jaus et al.· 0 citations
Federated learning (FL) enables multiple clinical institutions to collaboratively train a shared disease classifier without centralizing patient data. In practice, however, each institution annotates only the pathologies within its area of expertise, so the federation operates under task heterogeneity: each client holds labels for a strict subset of the target disease categories while the remaining classes are entirely unobserved at that site. Existing gradient-based FL methods fail under this setting because they require hundreds of communication rounds to converge and because missing class labels introduce systematic false-negative bias that the model cannot correct without a principled mechanism. We propose an analytic federated learning framework for multi-label medical image classification under task heterogeneity. The proposed method replaces iterative gradient optimization with three closed-form operations: a balanced label projection that neutralizes class-imbalance bias by normalizing positive and negative contributions to equal total mass; a per-class absolute aggregation law that independently assembles the optimal ridge-regression classifier for each disease category from the sufficient statistics uploaded by its annotating clients; and an optional analytic pseudo-label refinement round that propagates missingclass knowledge from a confidence-filtered teacher classifier to non-annotating clients. The entire procedure requires at most two communication rounds, irrespective of the degree of task heterogeneity or the number of participating clients. Experiments on ChestXray14 under four progressively severe missing-class configurations demonstrate that the proposed method consistently outperforms the state-of-the-art federated multi-label method FedMLP by up to 18.44 BACC points and 13.24 AUC points, while reducing the communication. Further experiments across different backbone architectures confirm that the proposed method is backbone-agnostic and generalizes across ResNet, VGG, and EfficientNet encoders without any hyperparameter adjustment.
Afsaneh Mahanipour, Hana Khamfroush· IEEE International Conferenc...· 0 citations
Efficiently identifying and retrieving imaging studies to create research cohorts is often hindered by the segregated storage of medical imaging data and radiological reports. To address this, we developed PACScrawler, an open-source tool designed to streamline cohort assembly by enabling efficient search and access to imaging studies across Picture Archiving and Communication Systems (PACS) and Radiology Information Systems (RIS). The system features a backend crawler module that regularly collects DICOM and RIS metadata and radiology reports from newly acquired studies and stores them in a local Apache Solr/Lucene index. A web-based interface enables multi-parameter search across the indexed database, and a retrieval module downloads the corresponding imaging data on demand. Between 2012 and 2024, PACScrawler indexed 2,327,317 examinations from 316,310 patients at our institution. In our local setup, routine indexing of 1079 examinations on a representative day required 26.5 min, corresponding to a median of 1.4 s per study. In a batch retrieval benchmark, ten head MRI examinations comprising 325 series were retrieved in 16 min 08 s using PACScrawler compared with 31 min 45 s by manual sequential PACS export. For LLM-assisted query construction, 17 of 20 generated Solr queries were usable with no or limited manual adjustment. As an illustrative use case, the index enabled longitudinal monitoring of local imaging volumes and examination complexity. PACScrawler lowers the technical barrier for clinical imaging research by providing an open-source, regularly updated local search index for selected PACS/RIS metadata and radiology reports. It supports retrospective cohort identification, on-demand retrieval of imaging data, and longitudinal monitoring of institutional imaging trends while routine indexing is limited to metadata and report text. As a DICOM-based tool, PACScrawler may be adapted to other clinical environments after local PACS/RIS integration and validation; source code is available at https://github.com/pacs-ris-crawler/pacs-ris-crawler .
Joshy Cyriac, A. Indrakanti, J. Wasserthal et al.· Journal of imaging informati...· 0 citations
