Retrieval-Guided Generation for Safer Histopathology Image Captioning
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866917454037385216 |
|---|---|
| author | Hoq, Md. Enamul Uegami, Wataru Alfasly, Saghir Alabtah, Ghazal Malakshan, Sahar Rahimi Kazemi, Armita Schmitgen, Alex T. Prior, Fred Tizhoosh, H. R. |
| author_facet | Hoq, Md. Enamul Uegami, Wataru Alfasly, Saghir Alabtah, Ghazal Malakshan, Sahar Rahimi Kazemi, Armita Schmitgen, Alex T. Prior, Fred Tizhoosh, H. R. |
| contents | Generative vision-language models can produce fluent medical image captions but remain prone to hallucination, over-specific diagnostic claims, and factual inconsistency-serious issues in pathology. We investigate retrieval-guided generation (RGG) as a safer alternative, where captions are formed by summarizing expert text from visually similar cases rather than generated de novo. On the ARCH histopathology dataset, RGG improves semantic alignment with ground truth, achieving cosine similarity of $\approx$0.60 versus $\approx$0.47 from MedGemma, with non-overlapping confidence intervals indicating a robust gain. A pathologist-led qualitative review shows better preservation of morphology-relevant terminology and fewer unsupported diagnoses, while revealing failure modes such as concept mixing and inherited over-specific labeling. Overall, retrieval-guided captioning offers a more transparent and reliable approach with clearer opportunities for auditing than fully generative methods. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_00893 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Retrieval-Guided Generation for Safer Histopathology Image Captioning Hoq, Md. Enamul Uegami, Wataru Alfasly, Saghir Alabtah, Ghazal Malakshan, Sahar Rahimi Kazemi, Armita Schmitgen, Alex T. Prior, Fred Tizhoosh, H. R. Computer Vision and Pattern Recognition Artificial Intelligence Information Retrieval Generative vision-language models can produce fluent medical image captions but remain prone to hallucination, over-specific diagnostic claims, and factual inconsistency-serious issues in pathology. We investigate retrieval-guided generation (RGG) as a safer alternative, where captions are formed by summarizing expert text from visually similar cases rather than generated de novo. On the ARCH histopathology dataset, RGG improves semantic alignment with ground truth, achieving cosine similarity of $\approx$0.60 versus $\approx$0.47 from MedGemma, with non-overlapping confidence intervals indicating a robust gain. A pathologist-led qualitative review shows better preservation of morphology-relevant terminology and fewer unsupported diagnoses, while revealing failure modes such as concept mixing and inherited over-specific labeling. Overall, retrieval-guided captioning offers a more transparent and reliable approach with clearer opportunities for auditing than fully generative methods. |
| title | Retrieval-Guided Generation for Safer Histopathology Image Captioning |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence Information Retrieval |
| url | https://arxiv.org/abs/2605.00893 |