CAP-IQA: Context-Aware Prompt-Guided CT Image Quality Assessment

TL;DR

CAP-IQA tackles CT image quality assessment by integrating medical text priors with image-specific context prompts and a causal debiasing strategy to separate idealized knowledge from actual degradations. The method combines a CNN-based visual encoder with a text encoder and employs Dynamic Cross-Prompt Attention to fuse semantic priors with image features, producing no-reference quality scores on a calibrated scale. On the LDCTIQA benchmark, CAP-IQA achieves the highest aggregate correlation $s = |r| + | ho| + | au|$ of $2.8590$, outperforming top competitors and demonstrating strong generalization to a large pediatric dataset. These results indicate CAP-IQA’s potential for robust, clinically aligned CT IQA across diverse imaging conditions and patient populations.

Abstract

Prompt-based methods, which encode medical priors through descriptive text, have been only minimally explored for CT Image Quality Assessment (IQA). While such prompts can embed prior knowledge about diagnostic quality, they often introduce bias by reflecting idealized definitions that may not hold under real-world degradations such as noise, motion artifacts, or scanner variability. To address this, we propose the Context-Aware Prompt-guided Image Quality Assessment (CAP-IQA) framework, which integrates text-level priors with instance-level context prompts and applies causal debiasing to separate idealized knowledge from factual, image-specific degradations. Our framework combines a CNN-based visual encoder with a domain-specific text encoder to assess diagnostic visibility, anatomical clarity, and noise perception in abdominal CT images. The model leverages radiology-style prompts and context-aware fusion to align semantic and perceptual representations. On the 2023 LDCTIQA challenge benchmark, CAP-IQA achieves an overall correlation score of 2.8590 (sum of PLCC, SROCC, and KROCC), surpassing the top-ranked leaderboard team (2.7427) by 4.24%. Moreover, our comprehensive ablation experiments confirm that prompt-guided fusion and the simplified encoder-only design jointly enhance feature alignment and interpretability. Furthermore, evaluation on an in-house dataset of 91,514 pediatric CT images demonstrates the true generalizability of CAP-IQA in assessing perceptual fidelity in a different patient population.

Figures (9)

• Figure 1: Our proposed CAP-IQA outperforms the top-ranked and recent models on the LDCTIQA 2023 benchmark lee2025low, achieving the highest correlation scores across Pearson’s linear correlation coefficient (PLCC), Spearman’s rank correlation coefficient (SROCC), and Kendall’s rank correlation coefficient (KROCC).
• Figure 2: Overview of the proposed CAP framework. The textual branch encodes medical priors, and the context branch extracts image-specific prompts from CT images. Both are fused through cross-prompt attention, aligning medical knowledge with visual features to reduce non-relevant biases and improve CT image quality assessment.
• Figure 3: Effectiveness of our CAP-IQA model in accurately assessing the quality of abdominal CT images across diverse IQA scores. Model predictions are in good agreement with the [actual] scores. CT images are visualized after removing MATLAB’s +1024 HU offset and re-windowing to Window Width: 400 and Window Level: 50.
• Figure 4: Kernel density plot across IQA score groups. Predictions remain centered near zero across all groups, indicating minimal bias of the model across different image quality levels.
• Figure 5: Generated text prompt description from Gemini for context-aware CT image quality assessment, including diagnostic visibility, interpretability, and artifact severity. The JSON structure for scores 0-4 is also presented.