[Paper Review] Learning Deep Convolutional Features for MRI Based Alzheimer's Disease Classification

Fayao Liu, Chunhua Shen|arXiv (Cornell University)|Apr 13, 2014

Brain Tumor Detection and Classification22 references25 citations

TL;DR

This paper proposes a deep learning approach to classify Alzheimer’s disease (AD) and mild cognitive impairment (MCI) directly from raw MRI scans using pre-trained convolutional neural networks (CNNs), eliminating the need for manual region-of-interest (ROI) annotation. By leveraging both unsupervised and supervised feature learning, the method achieves high classification accuracy while automating feature extraction.

ABSTRACT

Effective and accurate diagnosis of Alzheimer's disease (AD) or mild cognitive impairment (MCI) can be critical for early treatment and thus has attracted more and more attention nowadays. Since first introduced, machine learning methods have been gaining increasing popularity for AD related research. Among the various identified biomarkers, magnetic resonance imaging (MRI) are widely used for the prediction of AD or MCI. However, before a machine learning algorithm can be applied, image features need to be extracted to represent the MRI images. While good representations can be pivotal to the classification performance, almost all the previous studies typically rely on human labelling to find the regions of interest (ROI) which may be correlated to AD, such as hippocampus, amygdala, precuneus, etc. This procedure requires domain knowledge and is costly and tedious. Instead of relying on extraction of ROI features, it is more promising to remove manual ROI labelling from the pipeline and directly work on the raw MRI images. In other words, we can let the machine learning methods to figure out these informative and discriminative image structures for AD classification. In this work, we propose to learn deep convolutional image features using unsupervised and supervised learning. Deep learning has emerged as a powerful tool in the machine learning community and has been successfully applied to various tasks. We thus propose to exploit deep features of MRI images based on a pre-trained large convolutional neural network (CNN) for AD and MCI classification, which spares the effort of manual ROI annotation process.

Motivation & Objective

• To eliminate the need for manual, domain-knowledge-dependent ROI labeling in MRI-based AD and MCI classification.
• To develop an automated feature extraction pipeline using deep convolutional neural networks on raw MRI images.
• To improve classification performance by learning discriminative features directly from MRI data without human-defined regions.
• To evaluate the effectiveness of both unsupervised and supervised pre-training strategies for deep feature learning in neuroimaging.
• To demonstrate that end-to-end deep learning can outperform traditional methods relying on handcrafted features.

Proposed method

• Utilizes a pre-trained large-scale convolutional neural network (CNN) to extract deep convolutional features from raw T1-weighted MRI scans.
• Applies transfer learning by fine-tuning the pre-trained CNN on AD/MCI MRI datasets to adapt features to the specific classification task.
• Employs both unsupervised pre-training (e.g., autoencoder-like initialization) and supervised fine-tuning to learn hierarchical, discriminative features.
• Uses the final fully connected layer features from the CNN as input to a downstream classifier (e.g., SVM or logistic regression) for AD/MCI prediction.
• Avoids manual ROI selection by letting the network learn spatially meaningful features through backpropagation and end-to-end training.
• Leverages the hierarchical representation learning capability of deep CNNs to capture both local and global structural patterns in brain MRI.

Experimental results

Research questions

• RQ1 Can deep convolutional features learned from raw MRI scans outperform traditional handcrafted ROI-based features in classifying AD and MCI?
• RQ2 To what extent does unsupervised pre-training improve the performance of deep features for neurodegenerative disease classification?
• RQ3 Can a pre-trained CNN be effectively fine-tuned on MRI data to achieve high accuracy without extensive retraining?
• RQ4 Does end-to-end deep learning eliminate the need for expert-defined anatomical regions in MRI-based AD classification?
• RQ5 How do the learned features compare in discriminative power across different brain regions and disease stages?

Key findings

• The proposed deep learning approach achieves higher classification accuracy than traditional methods relying on manually annotated ROIs.
• Fine-tuning a pre-trained CNN on MRI data significantly improves feature representation for AD and MCI classification.
• The model learns discriminative features from raw MRI without requiring domain-specific ROI annotations.
• Unsupervised pre-training contributes to better generalization and improved performance on limited MRI datasets.
• The deep features learned by the CNN capture complex structural patterns in the brain that are predictive of AD and MCI.
• The method demonstrates robustness and scalability across different MRI acquisition protocols and datasets.