Persistence Image from 3D Medical Image: Superpixel and Optimized Gaussian Coefficient.

Area of Science:

• Medical Imaging
• Computational Topology
• Machine Learning

Background:

• Topological data analysis (TDA) using persistent homology reveals features missed by traditional deep learning in medical images.
• Existing TDA research predominantly focuses on 2D data, overlooking the full 3D context.

Purpose of the Study:

• To develop an innovative 3D TDA approach for comprehensive analysis of volumetric medical data.
• To enhance the modeling of 3D persistent homology for classification tasks.

Main Methods:

• A novel 3D TDA method integrating superpixels to convert 3D image features into point cloud data.
• Utilization of Optimized Gaussian Coefficient for efficient generation of 3D Persistence Images.
• Application to the MedMNist3D dataset for classification.

Main Results:

• The proposed 3D TDA method demonstrates superior performance compared to traditional approaches on the MedMNist3D dataset.
• Successfully generates holistic Persistence Images for 3D volumetric data.

Conclusions:

• The developed 3D TDA method shows significant potential for 3D persistent homology-based topological analysis in medical imaging classification.
• This approach effectively captures crucial topological properties in 3D medical data.