A novel multi-relation regularization method for regression and classification in AD diagnosis.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Alzheimer's disease diagnosis requires accurate prediction of clinical scores and disease status.
• Existing methods may not fully leverage the complex relationships within patient data.

Purpose of the Study:

• To develop a novel multi-relation regularization method for joint regression and classification in Alzheimer's disease diagnosis.
• To enhance feature selection and improve prediction accuracy for cognitive scores and disease classification.

Main Methods:

• A multi-relation regularization technique was proposed, incorporating feature-feature, response-response, and sample-sample relationships.
• An L2,1-norm was integrated within a least squares regression framework for effective feature selection.
• Support vector regression and classification models were trained for predicting ADAS-Cog, MMSE scores, and clinical labels.

Main Results:

• The proposed method demonstrated superior performance compared to state-of-the-art approaches on the Alzheimer's Disease Neuroimaging Initiative dataset.
• Significant improvements were observed in correlation coefficient and root mean squared error for regression tasks.
• Enhanced accuracy, sensitivity, specificity, and area under the ROC curve were achieved in classification.

Conclusions:

• The novel multi-relation regularization method effectively integrates diverse relational information for improved Alzheimer's disease diagnosis.
• This approach offers a promising tool for joint prediction of clinical scores and disease status, outperforming existing methods.