Search

Chest X-ray report generation is time-consuming and contributes to radiologist workload and burnout, motivating the need for AI systems that can reduce cognitive burden while preserving clinical accuracy. Although encoder-decoder models can generate reports from images, they often suffer from hallucinations, producing findings that are not present or missing real abnormalities due to lack of explicit grounding in evidence, making them unreliable for clinical use. To address this, we propose a cross-modal retrieval framework that generates reports by retrieving and assembling clinically validated sentences from existing radiology reports rather than generating text from scratch. The system uses contrastive learning to align chest X-ray image patches with report sentences in a shared embedding space, enabling retrieval of the most relevant clinical descriptions. A patch encoder extracts visual features, a sentence encoder represents report text, and a retrieval module identifies semantically matching sentences, which are then composed into a coherent final report. Because all outputs are sourced from real clinical reports, the method substantially reduces hallucinations while improving factual reliability and interpretability. This retrieval-based approach offers a scalable and safer alternative to generative models and can be evaluated on datasets such as MIMIC-CXR and CheXpert for clinical accuracy and retrieval performance.

Chest X-ray is a commonly used imaging tool in both acute and routine care, but the increasing reporting workload highlights the need for structured preliminary reports that aid triage, reduce delays, and ensure clinical relevance. Current AI systems often focus on classification or generic report generation, neglecting critical factors like free-text radiology requests, clinical history, and comparison context, leading to reports that, while technically fluent, are insufficiently focused. This article proposes a multimodal vision-language model that interprets both chest X-ray images and free-text radiology requests to generate structured preliminary reports directly addressing the clinical question. The model combines a radiographic encoder based on vision transformers, a text encoder for requests and prior reports, a cross-modal attention module, and a structured report decoder, organizing the output into relevant sections such as indication, technique, findings, impression, comparison, and answer-to-request. By aligning report generation with the clinical request, the model ensures that it answers specific questions—such as concerns about pneumonia, pulmonary oedema, or pneumothorax—improving report relevance, reducing misinterpretation, and supporting safer human-in-the-loop review. However, its effectiveness relies on accurate alignment, factual consistency, uncertainty management, and validation in real-world radiology settings.