Beyond the Embedding Bottleneck: Adaptive Retrieval-Augmented 3D CT Report Generation
This addresses the problem of generating more clinically effective radiology reports from 3D CT scans for medical professionals, representing an incremental improvement with specific gains.
The paper tackled the problem of incomplete pathology coverage in automated radiology report generation from 3D CT volumes by identifying a representational bottleneck in contrastive 3D CT embeddings, which exhibit severe dimensional concentration with as few as 2 effective dimensions out of 512. They proposed AdaRAG-CT, an adaptive augmentation framework that improves Clinical F1 from 0.420 to 0.480 (+6 points) on the CT-RATE benchmark.
Automated radiology report generation from 3D CT volumes often suffers from incomplete pathology coverage. We provide empirical evidence that this limitation stems from a representational bottleneck: contrastive 3D CT embeddings encode discriminative pathology signals, yet exhibit severe dimensional concentration, with as few as 2 effective dimensions out of 512. Corroborating this, scaling the language model yields no measurable improvement, suggesting that the bottleneck lies in the visual representation rather than the generator. This bottleneck limits both generation and retrieval; naive static retrieval fails to improve clinical efficacy and can even degrade performance. We propose \textbf{AdaRAG-CT}, an adaptive augmentation framework that compensates for this visual bottleneck by introducing supplementary textual information through controlled retrieval and selectively integrating it during generation. On the CT-RATE benchmark, AdaRAG-CT achieves state-of-the-art clinical efficacy, improving Clinical F1 from 0.420 (CT-Agent) to 0.480 (+6 points); ablation studies confirm that both the retrieval and generation components contribute to the improvement. Code is available at https://github.com/renjie-liang/Adaptive-RAG-for-3DCT-Report-Generation.