Multimodal and Multiscale Deep Neural Networks for the Early Diagnosis of Alzheimer's Disease using structural MR and FDG-PET images (1710.04782v1)

Abstract: Alzheimer's Disease (AD) is a progressive neurodegenerative disease. Amnestic mild cognitive impairment (MCI) is a common first symptom before the conversion to clinical impairment where the individual becomes unable to perform activities of daily living independently. Although there is currently no treatment available, the earlier a conclusive diagnosis is made, the earlier the potential for interventions to delay or perhaps even prevent progression to full-blown AD. Neuroimaging scans acquired from MRI and metabolism images obtained by FDG-PET provide in-vivo view into the structure and function (glucose metabolism) of the living brain. It is hypothesized that combining different image modalities could better characterize the change of human brain and result in a more accuracy early diagnosis of AD. In this paper, we proposed a novel framework to discriminate normal control(NC) subjects from subjects with AD pathology (AD and NC, MCI subjects convert to AD in future). Our novel approach utilizing a multimodal and multiscale deep neural network was found to deliver a 85.68\% accuracy in the prediction of subjects within 3 years to conversion. Cross validation experiments proved that it has better discrimination ability compared with results in existing published literature.

• The paper introduces a novel MMDNN framework that integrates multiscale MRI and FDG-PET imaging for early Alzheimer’s detection.
• It employs six parallel deep neural networks to process segmented patches, significantly outperforming previous state-of-the-art techniques.
• The model achieved an 85.68% accuracy in predicting conversion from MCI to AD, demonstrating promising clinical applications.

Multimodal and Multiscale Deep Neural Networks for the Early Diagnosis of Alzheimer's Disease

The paper presents a sophisticated framework leveraging deep learning methodologies to enhance the early diagnosis of Alzheimer’s Disease (AD) by integrating structural MRI and FDG-PET images. This paper mainly focuses on distinguishing between normal control (NC) subjects and those with Alzheimer's pathology, which also includes individuals with mild cognitive impairment (MCI) who are likely to progress to AD.

Methodology Overview

The proposed method employs what the authors refer to as a Multimodal Multiscale Deep Neural Network (MMDNN). This approach involves two key steps:

1. Image Preprocessing: Both MRI scans and FDG-PET images are segmented into patches. For MRI images, anatomical regions of interest (ROI) are demarcated, and patches are extracted based on voxel-wise clustering. These patches serve as feature vectors that represent the structural and metabolic activity of the brain.
2. Deep Neural Network Architecture: The MMDNN comprises six parallel deep neural networks, each processing a different scale of features from either MRI or FDG-PET data. These networks are then fused into a higher-level DNN to integrate and classify features from both imaging modalities.

Experimental Results

Impressively, the proposed model achieved an 85.68% accuracy in predicting conversion from MCI to AD within three years, surpassing previous models utilizing similar datasets. Various experiments were conducted to validate the diagnostic capability of this approach:

• Comparison with State-of-the-Art Techniques: Stressing the efficacy of their technique, they outperform several existing methods. Specifically, the paper demonstrates superior classification accuracy in distinguishing progressive MCI from stable MCI subjects using a combination of MRI and FDG-PET images.
• Multimodal and Multiscale Processing: Emphasizing features at different resolutions, the research shows that integrating multiscale data improves classification accuracy. Furthermore, combined MRI and FDG-PET modalities yield better discriminative performance than using each one alone.

Implications and Future Work

This research holds potential implications for both clinical practice and the theoretical expansion of deep learning applications in neuroimaging. Clinically, the early and accurate identification of individuals likely to progress to AD could inform treatment decisions and improve patient outcomes. Theoretically, this paper provides an exemplar of how neural networks can be effectively designed to handle multimodal data, expanding their application scope in medical diagnostics.

Moving forward, researchers might focus on integrating additional biomarkers, such as genetic data or cognitive assessments, to further enhance diagnostic accuracy. Moreover, adapting the model to newer, more expansive datasets could refine its predictive capabilities and generalizability.

Overall, this paper demonstrates noteworthy advances in applying deep neural networks to complex, multimodal medical data, presenting promising avenues for the early detection of neurodegenerative diseases such as Alzheimer's.