Left-right relationship-aware 3D volume classification method.

Area of Science:

• Medical imaging analysis
• Computational anatomy
• Machine learning in healthcare

Background:

• Bilateral symmetry (LR relationship) is crucial for understanding human anatomy and detecting abnormalities.
• Analyzing differences and similarities between left and right anatomical structures aids in medical diagnosis.
• Existing models may struggle with variations in body position and individual anatomical structures in 3D volumes.

Purpose of the Study:

• To propose a left-right (LR) relationship-aware classification model for 3D volumetric images.
• To enhance the detection of abnormalities by analyzing bilateral symmetry in anatomical structures.
• To improve diagnostic accuracy through a deeper understanding of anatomical structures.

Main Methods:

• Developed an LR relationship-aware classification model for 3D volumes.
• Employed an image feature extraction process focusing on LR symmetric positions.
• Introduced a multi-shift symmetric feature extraction module to handle positional variations in anatomy.

Main Results:

• The model demonstrated superior performance in 3D volume classification tasks for both lung and brain datasets.
• Experimental results confirmed the model's effectiveness in classifying structures with bilateral or semi-symmetric properties.
• The proposed method achieved better results compared to previous models in lung and brain classification.

Conclusions:

• The proposed LR relationship-aware classification model for 3D volumes effectively utilizes bilateral symmetry.
• The multi-shift symmetric feature extraction module successfully addresses positional gaps in LR anatomy.
• The model shows generalized performance and superior accuracy in classifying medical images with symmetric structures.