Centered Kernel Alignment Enhancing Neural Network Pretraining for MRI-Based Dementia Diagnosis.

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Dementia diagnosis, particularly for Alzheimer's disease (AD) and mild cognitive impairment (MCI), is crucial for elderly populations worldwide.
• Computer-aided diagnosis using resonance imaging (RI) shows promise but faces challenges due to the heterogeneous nature of MCI.
• Accurate discrimination between AD, MCI, and healthy controls (NC) is essential for timely medical intervention.

Purpose of the Study:

• To develop a novel supervised pretraining approach for artificial neural networks (ANNs) to enhance automated dementia diagnosis.
• To improve the classification accuracy and reduce class biasing in differentiating between MCI, AD, and NC using RI scans.
• To create more discriminating feature spaces for complex classification tasks in neurodegenerative disease diagnosis.

Main Methods:

• Introduced a supervised pretraining method for ANNs, initializing them with linear projections.
• Utilized the centered kernel alignment criterion to estimate projections, maximizing affinity between RI data and label matrices.
• Compared the proposed method against unsupervised initialization techniques (autoencoders, PCA) and top-performing CADDementia challenge classifiers.

Main Results:

• The proposed supervised pretraining approach significantly outperformed all baseline methods.
• Achieved a 7% improvement in classification accuracy and area under the receiver-operating-characteristic curve (AUC).
• Demonstrated a reduction in class biasing, leading to more balanced and reliable diagnostic outcomes.

Conclusions:

• The novel supervised pretraining method offers a more accurate and less biased approach to automated dementia diagnosis.
• This ANN-based strategy effectively utilizes RI data for improved discrimination of neurodegenerative conditions like MCI and AD.
• The findings suggest a promising direction for advancing computer-aided diagnostic tools in neurology.