arxivcs.CV2026-07-08
Seeing What Matters: Lesion-Aware High-Resolution Patch Discovery and Fusion for Chest X-ray Report Generation
Yingshu Li, Yunyi Liu, Zhenghao Chen, Tong Chen, Zailong Chen, Lingqiao Liu, et al.
Despite rapid advances in chest X-ray (CXR) foundation models, most radiology report generation (RRG) systems still rely on heavily downsampled inputs (e.g., 256x256) due to the fixed visual token budgets of pretrained vision encoders, suppressing subtle yet clinically important…