Organizations: University of Lagos · Machine Learning Collective · NITHUB, University of Lagos · Ahmadu Bello University · Federal University of Agriculture, Abeokuta (FUNAAB)
Retrieval-augmented generation is an attractive way to improve chest X-ray reporting, because reports from similar prior studies supply clinical context a general vision-language model lacks. We show the same mechanism is a reliable source of clinical error. Over 100 MIMIC-CXR studies with a frozen LLaVA-1.5 generator and BioMedCLIP retrieval, CheXbert clinical F1 doubles under relevant retrieval (0.201 to 0.402) and collapses to 0.043 under clinically mismatched retrieval, a fifth of the image-only score; the retrieval-induced hallucination rate, counting only unsupported findings traceable to retrieved evidence, rises from 0.00 to 0.76 and 0.98. To show this is not an artefact of evidence-set coverage, we introduce a coincidental-overlap control that scores image-only generations against evidence they never saw, placing the chance base rate at 0.18, four to five times below the observed rates. The generator does not merely acquire findings, it transcribes text: 95% of reports produced under relevant retrieval contain an eight-word span occurring verbatim in the retrieved evidence but absent from the reference, against 0% without retrieval. We then explain the mechanism: a normal chest X-ray retrieves at least one abnormal precedent in 24 of 28 cases, because medical image-embedding similarity is dominated by anatomy and acquisition rather than by the presence of disease. This has a direct design consequence. Retrieval similarity does not predict harm (RIH rates 0.80/0.80/0.60/0.84 across similarity quartiles), so relevance gates conditioned on embedding similarity cannot work; gating on predicted pathology agreement removes 61% of unsupported evidence at no cost to useful coverage. We argue that retrieval-augmented clinical systems must be evaluated under retrieval failure, not only under retrieval success.
Figures & tables
Metric
Retrieval
Mismatch
Label Jaccard@3
0.267
0.050
Label precision@3
0.272
0.000
Label recall@3 (abnormal targets)
0.604
0.000
≥1 shared pathology
0.55
0.00
Normal target → normal neighbours
4/28
5/28
Abnormal target →≥1 match
55/72
0/72
Table 1: BioMedCLIP top-3 retrieval quality over 100 evaluation studies.
Condition
n
Hall.
RIH
Copy
Omission
Clinical F1
No Retrieval
100
0.32
0.00
0.00
0.80
0.201
Prompt Retrieval
100
0.80
0.76
0.91
0.74
0.402
Mismatch Retrieval
100
0.98
0.98
1.00
0.78
0.043
Table 2: Results over 100 evaluation studies, labelled with CheXbert [ 12 ] . RIH: retrieval-induced hallucination. Copy: any finding shared by the generated report and retrieved evidence. Clinical F1 doubles under relevant retrieval and collapses under mismatch.
Gating signal
Kept
Unsup.
Coverage
Normal protected
No gate ( k=3 )
3.00
2.18
0.60
4/28
Cosine similarity, top-1
1.00
0.99
0.33
13/28
Label agreement, J≥0.2
1.42
0.84
0.60
28/28
Label agreement, J≥0.4
0.92
0.35
0.47
28/28
Label agreement, Lk⊆T
1.28
0.00
0.27
28/28
Table 3: Gating-signal comparison over 100 studies. “Unsup.”: mean findings in kept evidence but absent from the reference. “Coverage”: mean fraction of reference findings retained. “Normal protected”: normal-reference studies where no abnormal evidence survives. Label agreement is an oracle for the image side.
Large chest radiography archives are difficult to search because most studies are paired only with free-text reports rather than structured clinical annotations. Vision-language models offer a natural interface for text-to-image retrieval, but current biomedical models are primarily optimized for report-to-image matching rather than for satisfying short clinical search queries. This creates an objective mismatch: a model may retrieve images related to words in the query while failing to satisfy the full clinical constraint, especially for conjunctions and negations such as ``atelectasis and no pneumonia.'' We introduce CXR-Retrieve, a structured benchmark for compositional chest X-ray text-to-image retrieval. The benchmark contains 5,159 test images from the official test-split of MIMIC-CXR-JPG and 145 textual queries spanning single and conjunction findings, both positive and negative. Relevance is defined by whether a retrieved image satisfies all asserted pathology constraints, rather than by whether it matches a paired report. We further propose a label-aware contrastive fine-tuning objective for clinical retrieval. Our method attracts image-text pairs with compatible asserted pathology constraints, including shared confirmed absences, while explicitly repelling contradictory pairs. Starting from the in-domain CXR-CLIP checkpoint, our method improves Precision@5 over CXR-CLIP by 8.5 percentage points on two-pathology conjunctions and by 22.0 percentage points on negation queries. These results show that reliable chest X-ray retrieval requires training objectives that model not only which findings are mentioned, but also how they are clinically asserted.
Tomer Erez, Moshe Kimhi, Chaim Baskin +1
Technion – Israel Institute of Technology · Ben-Gurion University of the Negev
Automated radiology report generation has gained increasing attention with the rise of deep learning and large language models. However, fully generative approaches often suffer from hallucinations and lack clinical grounding, limiting their reliability in real-world workflows. In this study, we propose a multimodal retrieval-augmented generation (RAG) system for grounded drafting of chest radiograph impressions. The system combines contrastive image-text embeddings, case-based similarity retrieval, and citation-constrained draft generation to ensure factual alignment with historical radiology reports. A curated subset of the MIMIC-CXR dataset was used to construct a multimodal retrieval database. Image embeddings were generated using CLIP encoders, while textual embeddings were derived from structured impression sections. A fusion similarity framework was implemented using FAISS indexing for scalable nearest-neighbor retrieval. Retrieved cases were used to construct grounded prompts for draft impression generation, with safety mechanisms enforcing citation coverage and confidence-based refusal. Experimental results demonstrate that multimodal fusion significantly improves retrieval performance compared to image-only retrieval, achieving Recall@5 above 0.95 on clinically relevant findings. The grounded drafting pipeline produces interpretable outputs with explicit citation traceability, enabling improved trustworthiness compared to conventional generative approaches. This work highlights the potential of retrieval-augmented multimodal systems for reliable clinical decision support and radiology workflow augmentation
Vision-language models (VLMs) are increasingly used for medical image interpretation, yet they frequently hallucinate, generating clinically plausible but factually incorrect findings that pose direct patient safety risks. We introduce HalluCXR, a benchmark evaluating six architecturally diverse VLMs across 856 stratified MIMIC-CXR chest radiographs and three query types, yielding 15,408 model evaluations. An eight-category hallucination taxonomy with clinical severity ratings and a two-layer detection pipeline are validated against 250 human annotations (auto-detection F1=0.959; LLM judge F1=0.907). We find that 61.9--82.3% of outputs contain hallucinations, with clinically dangerous errors in up to 80.2%. Three key patterns emerge: normal radiographs paradoxically attract the most severe hallucinations, common findings are systematically over-fabricated while rare findings go under-detected, and response length alone predicts hallucination risk (AUC up to 0.908). A six-model ensemble reduces fabrication by up to 84.8% at the cost of increased omission; a three-model subset retains comparable performance at half the cost. These results establish that hallucination auditing, verbosity-based risk monitoring, and ensemble-based safety layers are prerequisites for clinical deployment.
Haoyu Wang, Zitong Li
Department of Biostatistics & Health Informatics, Institute of Psychiatry, Psychology & Neuroscience, King’s College London, London, UK