🤖 AI Summary
To address ambiguous, redundant, or incomplete abnormality descriptions in automated chest-abdominal CT report generation, this work proposes a “detect-then-generate” two-stage paradigm, wherein explicit abnormality detection serves as a clinically grounded guidance signal for lesion-driven, precise text generation. Methodologically, we design a unified framework integrating a 3D CNN-based abnormality localization module with an abnormality-conditioned Transformer report generator, incorporating an abnormality-masked attention mechanism and a hierarchical report decoding strategy. Evaluated on public benchmarks, our approach achieves a 12.6% improvement in BLEU-4 score and a 19.3% gain in clinical term accuracy over prior methods. Clinical evaluation by radiologists indicates that 92% of generated reports provide direct diagnostic utility. This work significantly enhances the clinical relevance and interpretability of automated reports, establishing a novel, pathology-aware paradigm for medical imaging report generation.
📝 Abstract
The rapid increase of computed tomography (CT) scans and their time-consuming manual analysis have created an urgent need for robust automated analysis techniques in clinical settings. These aim to assist radiologists and help them managing their growing workload. Existing methods typically generate entire reports directly from 3D CT images, without explicitly focusing on observed abnormalities. This unguided approach often results in repetitive content or incomplete reports, failing to prioritize anomaly-specific descriptions. We propose a new anomaly-guided report generation model, which first predicts abnormalities and then generates targeted descriptions for each. Evaluation on a public dataset demonstrates significant improvements in report quality and clinical relevance. We extend our work by conducting an ablation study to demonstrate its effectiveness.