Multi-modality canonical feature selection for Alzheimer's disease diagnosis

Area of Science:

• Medical Image Analysis
• Machine Learning
• Biostatistics

Background:

• High-dimensional medical imaging data presents challenges like the curse of dimensionality.
• Existing feature selection methods often overlook cross-modality feature correlations.

Purpose of the Study:

• To propose a novel multimodality canonical feature selection method.
• To enhance feature selection by incorporating cross-modality correlations.
• To improve clinical score prediction and disease identification in Alzheimer's disease.

Main Methods:

• Developed a multimodality canonical feature selection framework.
• Utilized canonical correlation analysis to project features into a canonical space.
• Employed sparse least squares regression with a novel canonical regularizer.

Main Results:

• The proposed method effectively selects canonical-cross-modality features.
• Demonstrated utility in predicting Alzheimer's Disease Assessment Scale-Cognitive subscale (ADAS-Cog) and Mini-Mental State Examination (MMSE) scores.
• Showed success in multiclass disease identification for Alzheimer's disease diagnosis.

Conclusions:

• The novel method advances feature selection for multimodality medical imaging.
• This approach offers improved performance for Alzheimer's disease clinical prediction and diagnosis.
• The technique effectively leverages complementary information from different imaging modalities.