Deep sequence modelling for Alzheimer's disease detection using MRI.

Area of Science:

• Artificial Intelligence in Medicine
• Neuroimaging Analysis
• Machine Learning for Healthcare

Background:

• Alzheimer's disease (AD) poses a significant health challenge, with early detection crucial for patient independence and treatment efficacy.
• Artificial intelligence (AI), particularly Convolutional Neural Networks (CNNs), shows promise for analyzing brain Magnetic Resonance Imaging (MRI) scans for early AD detection.
• Current 2D CNN approaches on 3D MRI data overlook temporal dependencies between image slices, limiting detection capabilities.

Purpose of the Study:

• To address the limitations of 2D CNNs in capturing temporal information from 3D MRI scans for Alzheimer's disease detection.
• To investigate the effectiveness of deep sequence-based networks in modeling MRI features for improved AD classification.

Main Methods:

• Utilized ResNet-18, a pre-trained CNN, to extract features from MRI scans.
• Implemented and compared deep sequence-based models, including Temporal Convolutional Networks (TCNs) and Recurrent Neural Networks (RNNs), to model feature sequences.
• Evaluated various configurations of these sequence-based models for Alzheimer's disease detection.

Main Results:

• The proposed Temporal Convolutional Network (TCN) model achieved superior classification performance.
• The TCN model demonstrated high accuracy (91.78%), sensitivity (91.56%), and specificity (92%).
• Sequence-based models enhanced the classification accuracy of 2D and 3D CNNs by up to 10%.

Conclusions:

• Modeling sequential MRI features with deep sequence-based networks significantly improves Alzheimer's disease detection.
• The TCN model represents a promising advancement in AI-driven early detection of Alzheimer's disease.
• This approach offers a substantial improvement over traditional 2D CNN methods for analyzing 3D MRI data.