Related Experiment Video
Updated: Jan 8, 2026
07:14
Dual Raster-Scanning Photoacoustic Small-Animal Imager for Vascular Visualization
Published on: July 15, 2020
4.5K
Hires-diagnoser: a dual stream medical image diagnosis framework based on multi-level resolution adaptive sensing
Si-Chao Zhao1, Jun-Jun Chen1, Shi-Long Shi1
1Guangdong Provincial Engineering and Technology Research Center of Light and Health, Guangdong Pharmaceutical University, Guangzhou 510006, People's Republic of China.
Biomedical Physics & Engineering Express
|December 11, 2025
Summary
Hires-Diagnoser, a dual-stream medical image diagnosis framework, enhances performance by integrating convolutional neural networks and transformers. This approach improves the detection of local and global features for accurate pathological interpretations.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Conventional convolutional neural networks (CNNs) struggle with global context due to fixed receptive fields.
- Transformers excel at long-range dependencies but have limitations in detecting small lesions.
- Integrating multi-scale features and spatial context is crucial for advanced medical image diagnosis.
Purpose of the Study:
- To introduce Hires-Diagnoser, a novel dual-stream framework for medical image diagnosis.
- To address limitations of existing models in capturing both local and global features across multiple resolutions.
- To improve the accuracy and adaptability of AI in medical image analysis.
Main Methods:
- A dual-stream architecture combining ConvNeXt (local texture features) and Swin-Transformer (global context) branches.
- Implementation of a cross-modal correlation (LCA) module for dynamic feature fusion across resolutions.
- Validation on diverse medical imaging datasets: RaabinWBC, Brain Tumor MRI, LC25000, and OCT-C8.
Main Results:
- Achieved high diagnostic accuracy rates: 99.45% (RaabinWBC), 98.01% (Brain Tumor MRI), 100% (LC25000), and 97.58% (OCT-C8).
- Demonstrated superior performance compared to existing state-of-the-art methods.
- Showcased effective integration of local texture and global contextual information.
Conclusions:
- Hires-Diagnoser provides a robust and adaptable solution for medical image diagnosis.
- The framework's cross-modal feature interaction mechanism enhances pathological interpretation capabilities.
- Significant potential for clinical application in diverse medical imaging scenarios.
