Shape-based disease grading via functional maps and graph convolutional networks with application to Alzheimer's...

Area of Science:

• Medical image analysis
• Computational anatomy
• Machine learning

Background:

• Traditional shape analysis methods struggle with incomplete or topologically varying anatomical data.
• Existing shape spaces have strict assumptions limiting their application.

Purpose of the Study:

• To adapt functional maps for analyzing complex anatomical shapes.
• To develop a graph-based learning approach for disease classification using novel shape descriptors.
• To improve the accuracy of differentiating Alzheimer's disease from normal controls using medical imaging data.

Main Methods:

• Adapted the concept of functional maps to overcome limitations of traditional shape analysis.
• Developed a graph-based learning algorithm incorporating new shape descriptors.
• Utilized the open-access Alzheimer's Disease Neuroimaging Initiative (ADNI) database for validation.

Main Results:

• The proposed method successfully analyzes incomplete and topologically varying shapes.
• The graph-based classifier demonstrated strong performance in differentiating Alzheimer's disease.
• The approach showed improved results compared to state-of-the-art geometric deep learning techniques.

Conclusions:

• Functional maps offer a flexible framework for advanced shape analysis in medical imaging.
• The developed graph-based learning method provides an effective tool for morphometric disease classification.
• This approach holds promise for improving early diagnosis and understanding of neurodegenerative diseases like Alzheimer's.