Fine-Grained Lesion Classification Framework for Early Auxiliary Diagnosis.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Computer Vision

Background:

• Deep neural networks are crucial for early disease diagnosis from medical images.
• Early-stage diseases present subtle visual differences between patient and healthy images, posing a Fine-Grained Visual Classification (FGVC) challenge.
• Existing FGVC methods struggle with variable lesion shapes/sizes and complex background relationships in medical imaging.

Purpose of the Study:

• To develop an advanced fine-grained lesion classification framework for early auxiliary diagnosis.
• To address the limitations of standard FGVC approaches in medical image analysis.
• To improve the accuracy of early disease detection through enhanced lesion identification and classification.

Main Methods:

• Proposed a novel fine-grained lesion classification framework.
• Implemented a method to accurately locate and extract multiple lesions of varying sizes and shapes.
• Utilized an attention mechanism to fuse lesion and background features effectively.

Main Results:

• The proposed model demonstrated accurate lesion localization capabilities.
• Experimental results on two real-world clinical datasets showed superior performance compared to existing methods.
• The framework effectively handles variability in lesion characteristics and background complexity.

Conclusions:

• The developed framework offers a significant advancement in early disease diagnosis using medical images.
• The attention-based fusion of lesion and background features enhances classification accuracy.
• This approach shows promise for improving clinical decision-making in early disease detection.