S2D-Align for Anatomically-Grounded RRG (AAAI 2026) by Liu ChangS2D-Align for Anatomically-Grounded RRG (AAAI 2026) by Liu Chang

S2D-Align for Anatomically-Grounded RRG (AAAI 2026)

Portfolio project
Liu Chang
Motivation & Challenge: Standard fine-tuning aligns radiographs and reports only at the instance level, giving weak anatomical grounding. S2D-Align enriches alignment progressively with auxiliary signals: coarse radiograph-report pairing, reference reports for instance-level guidance, and key phrases grounding generation in anatomical details, bridged by a memory-based adapter. Results & Venues: Achieved state-of-the-art results on MIMIC-CXR and IU X-Ray; ablations validated the multi-stage design. Published at AAAI 2026. Link: https://ojs.aaai.org/index.php/AAAI/article/download/40335/44296

1 like · 1 view · Sep 15, 2026

Motivation & Challenge: Standard fine-tuning aligns radiographs and reports only at the instance level, giving weak anatomical grounding.

Tags