Advancements in Radiology Report Generation and Lesion Segmentation

The field of medical imaging is witnessing significant advancements in radiology report generation and lesion segmentation. Researchers are focusing on developing novel approaches that incorporate clinical context, prior knowledge, and anatomical information to improve the accuracy and fluency of generated reports. The use of hierarchical graph attention networks, prior-guided contrastive pre-training, and joint temporal and clinical modeling are some of the innovative techniques being explored. These methods have shown promising results in capturing diagnostic intent, tracking disease progression, and segmenting small lesions in medical images. Noteworthy papers in this area include LesiOnTime, which proposes a 3D segmentation approach that leverages longitudinal imaging and BI-RADS scores, and R2GenKG, which constructs a large-scale multi-modal medical knowledge graph for LLM-based radiology report generation. PriorRG is also notable for its prior-guided contrastive pre-training scheme and coarse-to-fine decoding for report generation. CT-GRAPH is another significant contribution, which uses a hierarchical graph attention network to model radiological knowledge and generate detailed medical reports.

Sources

LesiOnTime -- Joint Temporal and Clinical Modeling for Small Breast Lesion Segmentation in Longitudinal DCE-MRI

R2GenKG: Hierarchical Multi-modal Knowledge Graph for LLM-based Radiology Report Generation

PriorRG: Prior-Guided Contrastive Pre-training and Coarse-to-Fine Decoding for Chest X-ray Report Generation

CT-GRAPH: Hierarchical Graph Attention Network for Anatomy-Guided CT Report Generation

Built with on top of