Multimodal neuroimage data fusion based on multikernel learning in personalized medicine.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Personalized Medicine

Background:

• Neuroimaging is crucial for diagnosing brain diseases.
• Artificial intelligence enhances neuroimaging analysis beyond traditional methods.
• Single neuroimaging modalities may miss critical diagnostic patterns.

Purpose of the Study:

• To propose a multikernel approach for fusing multimodal neuroimaging data.
• To improve diagnostic accuracy by exploring complementary patterns across different data types.
• To enhance the generalization ability of diagnostic models.

Main Methods:

• Developed a multikernel version of the regularized label relaxation linear regression model.
• Applied the model to fuse multimodal data for joint diagnosis.
• Evaluated the method using Alzheimer's disease diagnosis data.

Main Results:

• Multimodality fusion via multikernel learning outperformed single modality approaches.
• The proposed method effectively explores complementary patterns across different imaging types.
• Reduced differences between training and testing performance indicate improved generalization and reduced overfitting.

Conclusions:

• Multikernel learning offers a superior approach for multimodal neuroimaging data fusion.
• This method enhances diagnostic performance and generalization for brain diseases like Alzheimer's.
• The approach holds promise for advancing personalized medicine in neurodiagnostics.