Diagnosis of Alzheimer's Disease Based on Structural MRI Images Using a Regularized Extreme Learning Machine and PCA...

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder leading to dementia, with no current cure.
• Early diagnosis of AD is vital for patient care and research, but faces challenges due to limited training data and complex feature representations.
• Structural magnetic resonance (sMR) imaging offers potential for early AD detection.

Purpose of the Study:

• To compare the effectiveness of different machine learning classifiers for diagnosing AD, mild cognitive impairment (MCI), and healthy control (HC) subjects.
• To evaluate the performance of Support Vector Machine (SVM), Import Vector Machine (IVM), and Regularized Extreme Learning Machine (RELM) using sMR images.
• To assess the impact of feature selection and kernel-based approaches on classification accuracy.

Main Methods:

• Utilized volumetric sMR image data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.
• Employed a greedy score-based feature selection technique to identify critical feature vectors.
• Applied kernel-based discriminative approaches to handle intricate data distributions.
• Compared classification performance of SVM, IVM, and RELM algorithms.

Main Results:

• The Regularized Extreme Learning Machine (RELM) combined with the feature selection method demonstrated a significant enhancement in classification accuracy.
• The proposed approach effectively discriminated between Alzheimer's disease (AD), mild cognitive impairment (MCI), and healthy control (HC) subjects.
• Feature selection proved crucial for improving the diagnostic performance of the machine learning models.

Conclusions:

• The RELM classifier, augmented by a feature selection strategy, offers a promising and accurate method for the early diagnosis of Alzheimer's disease using sMR imaging.
• This approach addresses key challenges in AD diagnosis, including limited sample sizes and high dimensionality of features.
• The findings support the utility of advanced machine learning techniques in neuroimaging for neurodegenerative disease detection.