Novel deep learning for multi-class classification of Alzheimer's in disability using MRI datasets

Area of Science:

• Neuroimaging and Machine Learning
• Computational Neuroscience
• Medical Diagnostics

Background:

• Alzheimer's disease (AD) is a leading cause of neurodegeneration, characterized by cognitive and motor impairments.
• Accurate classification of AD subtypes from MRI scans is challenging due to overlapping features.
• Existing machine learning (ML) and deep learning (DL) methods struggle with precise multi-class AD identification.

Purpose of the Study:

• To develop and evaluate transfer learning-based feature extraction methods for multi-class classification of AD categories from MRI scans.
• To propose a novel deep learning model, 'IncepRes', by fusing Inception and ResNet architectures for enhanced AD classification.
• To investigate the performance of different transfer learning models (ResNet152V2, VGG16, InceptionV3, MobileNet) on AD datasets.

Main Methods:

• Employed four transfer learning models (ResNet152V2, VGG16, InceptionV3, MobileNet) for feature extraction from MRI scans.
• Utilized publicly available datasets: Alzheimer's Disease Neuroimaging Initiative (ADNI) and Open Access Series of Imaging Studies (OASIS), along with a merged dataset.
• Developed a hybrid Convolutional Neural Network (CNN) model, 'IncepRes', integrating ResNet152V2 and Inception architectures for multi-class AD classification.

Main Results:

• Modified ResNet152V2 demonstrated superior performance as a feature extractor among the evaluated transfer learning methods.
• The proposed 'IncepRes' model achieved high classification accuracies: 96.96% (ADNI), 98.35% (OASIS), and 97.13% (Merged dataset).
• The 'IncepRes' model outperformed other competing deep learning structures in classifying various Alzheimer's disease categories.

Conclusions:

• The proposed transfer learning framework and 'IncepRes' model enable automated and precise classification of Alzheimer's disease categories from MRI data.
• This advancement holds potential for timely management and treatment of cognitive and functional impairments associated with Alzheimer's disease.
• The study highlights the efficacy of deep learning in addressing the challenges of multi-class classification in neurodegenerative disease diagnosis.