Label-noise resistant logistic regression for functional data classification with an application to Alzheimer's...

Area of Science:

• Neuroimaging
• Biostatistics
• Machine Learning

Background:

• Alzheimer's disease (AD) diagnosis relies on cognitive tests, risking misdiagnosis.
• Structural changes in the corpus callosum (CC) during AD progression offer potential diagnostic markers.
• Misclassified labels in diagnostic datasets significantly impair classification model performance.

Purpose of the Study:

• To develop a novel logistic regression model for functional data classification that is robust to mislabeled diagnostic data.
• To enhance the accuracy of Alzheimer's disease classification by incorporating corpus callosum thickness.
• To identify potentially mislabeled observations within a dataset.

Main Methods:

• Proposed a functional logistic regression model with individual intercepts to address label noise.
• Employed an MM algorithm for efficient model training and parameter estimation.
• Validated the method using synthetic datasets and applied it to real MRI data for AD classification.

Main Results:

• The proposed model demonstrated superior performance compared to existing methods on synthetic datasets.
• The method successfully differentiated patients with Alzheimer's disease from healthy controls using corpus callosum MRI data.
• The individual intercepts effectively flagged potentially mislabeled observations.

Conclusions:

• The developed logistic regression model offers a robust and efficient approach for classifying Alzheimer's disease, particularly in the presence of noisy labels.
• Corpus callosum thickness derived from MRI is a valuable covariate for AD diagnosis.
• This method has the potential to improve diagnostic accuracy and reliability in clinical settings.