Beyond the Embedding Bottleneck: Adaptive Retrieval-Augmented 3D CT Report Generation
Automated radiology report generation from 3D CT scans often misses details due to limitations in how visual data is represented by AI embeddings. This work diagnoses a 'representational bottleneck' where visual features fail to capture fine-grained clinical semantics. The AdaRAG-CT framework solves this by adaptively retrieving relevant text context during generation, achieving state-of-the-art clinical accuracy.