arXiv:2507.17234v3 Announce Type: replace 
Abstract: Automatic generation of radiology reports has the potential to alleviate radiologists' significant workload, yet current methods struggle to deliver clinically reliable conclusions. In particular, most prior approaches focus on producing fluent text without effectively ensuring the factual correctness of the reports and often rely on single-view images, limiting diagnostic comprehensiveness. We propose CLARIFID, a novel framework that directly optimizes diagnostic correctness by mirroring the two-step workflow of experts. Specifically, CLARIFID (1) learns the logical flow from Findings to Impression through section-aware pretraining, (2) is fine-tuned with Proximal Policy Optimization in which the CheXbert F1 score of the Impression section serves as the reward, (3) employs controlled decoding that completes "Findings" before synthesizing the "Impression", and (4) fuses multiple chest X-ray views via a vision-transformer-based multi-view encoder. During inference, we apply a next-token forcing strategy followed by report-level re-ranking, ensuring that the model first produces a comprehensive "Findings" section before synthesizing the "Impression" and thereby preserving coherent clinical reasoning. Experimental results on the MIMIC-CXR dataset demonstrate that our method achieves superior clinical efficacy and outperforms existing baselines on clinical efficacy scores.

يمثل تقديم CLARIFID تقدمًا كبيرًا في توليد تقارير الأشعة تلقائيًا، حيث يتناول التحدي المتمثل في تقديم استنتاجات موثوقة سريريًا. يعمل هذا الإطار على تحسين دقة التشخيص من خلال محاكاة سير العمل لدى الخبراء، مما يعزز التدفق المنطقي من النتائج إلى الانطباعات ويستخدم عدة مشاهد للأشعة السينية للصدر لتحليل شامل.

La introducción de CLARIFID representa un avance significativo en la generación automática de informes de radiología, abordando el desafío de ofrecer conclusiones clínicamente fiables. Este marco optimiza la corrección diagnóstica al imitar el flujo de trabajo de los expertos, mejorando el flujo lógico de los hallazgos a las impresiones y utilizando múltiples vistas de radiografías de tórax para un análisis integral.

L'introduction de CLARIFID représente une avancée significative dans la génération automatique de rapports de radiologie, répondant au défi de fournir des conclusions cliniquement fiables. Ce cadre optimise la justesse diagnostique en imitant le flux de travail des experts, améliorant ainsi le flux logique des constatations aux impressions et utilisant plusieurs vues de radiographies thoraciques pour une analyse complète.

The introduction of CLARIFID represents a significant advancement in the automatic generation of radiology reports, addressing the challenge of delivering clinically reliable conclusions. This framework optimizes diagnostic correctness by mimicking the expert workflow, enhancing the logical flow from findings to impressions and utilizing multiple chest X-ray views for comprehensive analysis.

CLARIFID: Improving Radiology Report Generation by Reinforcing Clinically Accurate Impressions and Enforcing Detailed Findings

arXiv:2512.06014v1 Announce Type: new 
Abstract: Recent foundation models have demonstrated strong performance in medical image representation learning, yet their comparative behaviour across datasets remains underexplored. This work benchmarks two large-scale chest X-ray (CXR) embedding models (CXR-Foundation (ELIXR v2.0) and MedImagelnsight) on public MIMIC-CR and NIH ChestX-ray14 datasets. Each model was evaluated using a unified preprocessing pipeline and fixed downstream classifiers to ensure reproducible comparison. We extracted embeddings directly from pre-trained encoders, trained lightweight LightGBM classifiers on multiple disease labels, and reported mean AUROC, and F1-score with 95% confidence intervals. MedImageInsight achieved slightly higher performance across most tasks, while CXR-Foundation exhibited strong cross-dataset stability. Unsupervised clustering of MedImageIn-sight embeddings further revealed a coherent disease-specific structure consistent with quantitative results. The results highlight the need for standardised evaluation of medical foundation models and establish reproducible baselines for future multimodal and clinical integration studies.

أظهرت الأبحاث الأخيرة تقييمًا لنموذجين من نماذج تصوير الأشعة السينية للصدر (CXR)، وهما CXR-Foundation (ELIXR v2.0) وMedImageInsight، باستخدام مجموعات البيانات العامة MIMIC-CXR وNIH ChestX-ray14. استخدمت الدراسة خط أنابيب موحد للمعالجة المسبقة ومصنفات ثابتة لضمان القابلية للتكرار، مما يكشف أن MedImageInsight يتفوق عمومًا على CXR-Foundation، الذي يظهر استقرارًا قويًا عبر مجموعات البيانات.

Una investigación reciente evalúa dos modelos de incrustación de radiografías de tórax (CXR), CXR-Foundation (ELIXR v2.0) y MedImageInsight, utilizando los conjuntos de datos públicos MIMIC-CXR y NIH ChestX-ray14. El estudio emplea un pipeline de preprocesamiento unificado y clasificadores fijos para garantizar la reproducibilidad, revelando que MedImageInsight generalmente supera a CXR-Foundation, que muestra una fuerte estabilidad entre conjuntos de datos.

Une recherche récente évalue deux modèles d'embedding de radiographies thoraciques (CXR), CXR-Foundation (ELIXR v2.0) et MedImageInsight, en utilisant les ensembles de données MIMIC-CXR et NIH ChestX-ray14 disponibles publiquement. L'étude utilise un pipeline de prétraitement unifié et des classificateurs fixes pour garantir la reproductibilité, révélant que MedImageInsight surpasse généralement CXR-Foundation, qui montre une forte stabilité entre les ensembles de données.

Recent research benchmarks two chest X-ray (CXR) embedding models, CXR-Foundation (ELIXR v2.0) and MedImageInsight, using the publicly available MIMIC-CXR and NIH ChestX-ray14 datasets. The study employs a unified preprocessing pipeline and fixed classifiers to ensure reproducibility, revealing that MedImageInsight generally outperforms CXR-Foundation, which shows strong stability across datasets.

CLARIFID: Improving Radiology Report Generation by Reinforcing Clinically Accurate Impressions and Enforcing Detailed Findings

Was this article worth reading? Share it