Saved in:
Bibliographic Details
Main Authors: Wang, Chenyu, Dai, Weicheng, Liu, Han, Li, Wenchao, Batmanghelich, Kayhan
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.10437
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914466379071488
author Wang, Chenyu
Dai, Weicheng
Liu, Han
Li, Wenchao
Batmanghelich, Kayhan
author_facet Wang, Chenyu
Dai, Weicheng
Liu, Han
Li, Wenchao
Batmanghelich, Kayhan
contents Vision--language models (VLMs) for radiology report generation (RRG) can produce long-form chest CT reports from volumetric scans and show strong potential to improve radiology workflow efficiency and consistency. However, existing methods face two key limitations: (i) training supervision is often coarse, aligning a whole CT volume with a full free-text report without explicit alignment for fine-grained attributes or pathology locations; and (ii) evaluation is typically holistic (lexical overlap, entity matching, or LLM-as-a-judge scores) and not diagnostic for spatial grounding. We propose \emph{Discriminative Cue-Prompting with Prompt Dropout (DCP-PD)}, a plug-and-play framework that distills fine-grained cues from free-text reports and uses them to guide report generation while mitigating shortcut reliance via prompt dropout. DCP-PD achieves state-of-the-art performance on CT-RATE, improving macro F1 from $=0.501$ to $0.603$ (20% relative), and substantially boosts out-of-distribution performance on Rad-ChestCT from F1 $=0.266$ to $0.503$ (89% relative). Finally, we introduce a hierarchical, location-aware question-set protocol (presence $\rightarrow$ laterality $\rightarrow$ lobe) to directly assess pathology-location grounding, showing that fine-grained spatial localization remains challenging even for models that score highly on current benchmarks.
format Preprint
id arxiv_https___arxiv_org_abs_2604_10437
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Enhancing Fine-Grained Spatial Grounding in 3D CT Report Generation via Discriminative Guidance
Wang, Chenyu
Dai, Weicheng
Liu, Han
Li, Wenchao
Batmanghelich, Kayhan
Computer Vision and Pattern Recognition
Vision--language models (VLMs) for radiology report generation (RRG) can produce long-form chest CT reports from volumetric scans and show strong potential to improve radiology workflow efficiency and consistency. However, existing methods face two key limitations: (i) training supervision is often coarse, aligning a whole CT volume with a full free-text report without explicit alignment for fine-grained attributes or pathology locations; and (ii) evaluation is typically holistic (lexical overlap, entity matching, or LLM-as-a-judge scores) and not diagnostic for spatial grounding. We propose \emph{Discriminative Cue-Prompting with Prompt Dropout (DCP-PD)}, a plug-and-play framework that distills fine-grained cues from free-text reports and uses them to guide report generation while mitigating shortcut reliance via prompt dropout. DCP-PD achieves state-of-the-art performance on CT-RATE, improving macro F1 from $=0.501$ to $0.603$ (20% relative), and substantially boosts out-of-distribution performance on Rad-ChestCT from F1 $=0.266$ to $0.503$ (89% relative). Finally, we introduce a hierarchical, location-aware question-set protocol (presence $\rightarrow$ laterality $\rightarrow$ lobe) to directly assess pathology-location grounding, showing that fine-grained spatial localization remains challenging even for models that score highly on current benchmarks.
title Enhancing Fine-Grained Spatial Grounding in 3D CT Report Generation via Discriminative Guidance
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2604.10437