Radiology report generation based on adaptive enhanced fusion of multi features.
Journal:
Computers in biology and medicine
Published Date:
May 31, 2025
Abstract
The radiology report is essential for doctors' diagnosis and treatment. The automatic generation of radiology reports can assist doctors in diagnosis and treatment, thereby reducing their workload. Some existing studies consider the entire radiological image as the input for report generation, overlooking the local feature information within it. To leverage this local information, this paper introduces a novel framework, Stratified Adaptive Enhanced Net(SAENet), to adaptively fuse and enhance global and local features. Initially, SAENet segregates multi-layer features and merges different layers of features by connecting them horizontally. Subsequently, the fused features undergo a feature adaptive enhancement module to improve the feature representation. Finally, the caption generator produces radiological reports based on the fused and enhanced feature information. The experiments were conducted using the IU X-ray dataset and Peir Gross dataset. The experimental results demonstrate that the model achieves competitive results compared to many previous methods.