[Paper Review] 3D CNN-based classification using sMRI and MD-DTI images for Alzheimer disease studies

Alexander Khvostikov, Karim Aderghal+3 more|arXiv (Cornell University)|Jan 18, 2018

Dementia and Cognitive Impairment Research71 citations

TL;DR

The paper proposes a 3D CNN framework that fuses hippocampal ROI sMRI and MD-DTI (Mean Diffusivity) maps to classify AD, MCI, and NC, including ROI-size analysis and a dedicated data augmentation strategy to balance classes.

ABSTRACT

Computer-aided early diagnosis of Alzheimers Disease (AD) and its prodromal form, Mild Cognitive Impairment (MCI), has been the subject of extensive research in recent years. Some recent studies have shown promising results in the AD and MCI determination using structural and functional Magnetic Resonance Imaging (sMRI, fMRI), Positron Emission Tomography (PET) and Diffusion Tensor Imaging (DTI) modalities. Furthermore, fusion of imaging modalities in a supervised machine learning framework has shown promising direction of research. In this paper we first review major trends in automatic classification methods such as feature extraction based methods as well as deep learning approaches in medical image analysis applied to the field of Alzheimer's Disease diagnostics. Then we propose our own algorithm for Alzheimer's Disease diagnostics based on a convolutional neural network and sMRI and DTI modalities fusion on hippocampal ROI using data from the Alzheimers Disease Neuroimaging Initiative (ADNI) database (http://adni.loni.usc.edu). Comparison with a single modality approach shows promising results. We also propose our own method of data augmentation for balancing classes of different size and analyze the impact of the ROI size on the classification results as well.

Motivation & Objective

• Motivate and survey current automatic classification methods for Alzheimer’s disease and its prodromal stages (AD/MCI) using multimodal imaging.
• Propose a CNN-based framework that fuses sMRI and MD-DTI data from the hippocampus ROI for AD diagnostics.
• Evaluate the impact of ROI size and augmentation strategies on classification performance.
• Leverage ADNI data to demonstrate the feasibility of ROI-focused 3D CNN fusion in a limited-data scenario.

Proposed method

• Adapted 3D CNN architecture operating on hippocampal ROI volumes from sMRI and MD-DTI maps.
• Siamese-like late fusion: separate modality pipelines feed into a shared fully-connected layer before softmax.
• Data augmentation with controlled balancing: random shifts and Gaussian blur to generate additional samples for underrepresented classes.
• ROI sizes investigated include 28-, 38-, 42-, and 48-voxel cubes encasing the hippocampus ROI.
• Batch normalization and Nesterov momentum are used to stabilize and accelerate training.
• A training strategy inspired by cross-validation with repeated train/validation splits to maximize data utilization.

Experimental results

Research questions

• RQ1 Can fusion of sMRI and MD-DTI hippocampal ROI data via 3D CNNs improve AD/NC/MCI classification over single modalities?
• RQ2 How does ROI size influence classification accuracy in ROI-based 3D CNN architectures?
• RQ3 What is the effect of targeted data augmentation and class balancing on model performance in small-to-moderate ADNI-derived datasets?

Key findings

• The study shows that combining sMRI and MD-DTI within hippocampal ROIs yields competitive classification results compared to single-modality approaches.
• Larger ROIs extending beyond the base hippocampal region can influence performance, with different architectures yielding varying gains.
• A balanced augmentation strategy helps mitigate class-imbalance and improves generalization in the network training.
• The proposed 3D CNN architectures demonstrate robust training behavior through batch normalization and carefully chosen optimization settings.
• The approach demonstrates feasibility of ROI-based, multimodal 3D CNN classification on a subset of ADNI data.