Diagnosis and Prognosis of Alzheimer’s Disease Using Brain Morphometry and White Matter Connectomes

Yun Wang, Chenxiao Xu, Ji Hwan Park, Seonjoo Lee, Yaakov Stern, Shinjae Yoo, Jong Hun Kim, Hyoung Seop Kim, Jiook Cha, The Alzheimer’s Disease Neuroimaging InitiativebioRxiv 2018 · DOI

TL;DR

This paper explores using multimodal MRI data (structural and diffusion) combined with machine learning to diagnose Alzheimer's Disease (AD) and Mild Cognitive Impairment (MCI). They found that combining brain morphometry (structure) and white matter connectomics (connections) significantly improved diagnostic accuracy compared to traditional methods, and even showed promise in predicting disease progression.

What Problem Does This Paper Address?

Early and accurate diagnosis of Alzheimer's Disease (AD) is crucial for developing and administering effective treatments. While structural MRI (like T1-weighted scans) has been used to identify changes in brain volume and shape (morphometry), these markers alone often lack the accuracy and generalizability needed for widespread clinical use. Diffusion MRI (dMRI) offers a way to probe the brain's white matter "wiring" by tracking water diffusion, allowing for the reconstruction of structural connections between brain regions (connectomics). The challenge this paper addresses is whether combining these different types of MRI data, along with advanced machine learning, can lead to more robust and reliable diagnostic tools for AD and MCI, and even predict disease progression.

Methods

The researchers employed a machine learning framework to analyze multimodal MRI data. They extracted two main types of features:

  • Morphometry: This involved analyzing structural MRI (T1-weighted and FLAIR) to measure aspects like cortical thickness, surface area, and volumes of specific brain regions, particularly the hippocampus. They used established pipelines like Freesurfer for this.
  • White Matter Connectomics: Using diffusion MRI, they reconstructed the brain's white matter tracts. This involved advanced processing steps like denoising, eddy current correction, and probabilistic tractography to estimate the strength and patterns of connections between different brain regions. They generated a large number of features (33,698 per subject) representing these structural connections.

These features were then fed into machine learning classifiers, specifically random forest, logistic regression, and support vector machines. They used nested cross-validation to train and test their models, ensuring robust performance estimation and avoiding overfitting. To evaluate their approach, they compared their machine learning models against established "benchmark" models, including white matter hyperintensity (WMH) volumes from FLAIR MRI and cerebrospinal fluid (CSF) biomarkers.

Datasets

The study utilized two independent datasets:

  1. Korean National Health Insurance Service-Ilsan Hospital (NHIS-IH) Cohort: This dataset, comprising 211 participants (110 AD, 64 MCI, 37 Subjective Memory Complaints - SMC), was used for initial model building and testing. SMC individuals were considered cognitively normal.
  2. Alzheimer's Disease Neuroimaging Initiative (ADNI)-2 Cohort: This larger, multi-site dataset (179 participants) was used to test the generalizability of the developed models.

The NHIS-IH cohort served as the primary dataset for developing and validating the machine learning models, while the ADNI-2 dataset was used to confirm their performance across different populations and scanning sites.

Key Insights and Ideas

  • Connectomics Adds Value: The study strongly suggests that white matter connectomics, derived from diffusion MRI, provides crucial information for diagnosing AD and MCI that is complementary to traditional structural morphometry.
  • Synergy in Diagnosis: Combining morphometry and connectomics features often led to the highest diagnostic accuracies, indicating a synergistic effect where the whole is greater than the sum of its parts for classification.
  • Sensitivity to Early Changes: The connectome-based models showed a particular advantage in distinguishing between MCI and SMC (cognitively normal individuals with subjective complaints), suggesting a higher sensitivity to subtle white matter changes associated with early neurodegeneration.
  • Prognostic Potential: While morphometry alone showed the best performance in predicting MCI to AD progression, the study demonstrated that MRI-based methods, including connectomics, hold promise for predicting disease trajectory.

Key Results

  • High Diagnostic Accuracy: In the NHIS-IH cohort, the combined morphometry and connectome models achieved excellent classification accuracies: 97% for AD vs. SMC, 83% for MCI vs. SMC, and 97% for AD vs. MCI. These significantly outperformed the white matter hyperintensity benchmark.
  • Generalizability Confirmed: In the ADNI-2 dataset, the combined models also showed strong performance (e.g., 96% for AD vs. HC, 75% for AD vs. MCI), comparable to or better than CSF biomarkers, demonstrating the robustness of the approach across different datasets.
  • Predicting Progression: For predicting MCI to AD progression in ADNI-2, the morphometry model achieved 69% accuracy, similar to the CSF benchmark (70%), while the connectome and combined models showed meaningful, albeit slightly lower, accuracies.

Limitations and Future Work

  • Site Variability: The study noted a ~10% decrease in model performance in the multi-site ADNI-2 data compared to the single-site NHIS-IH data. This highlights the challenge of inter-site variability in dMRI data and the need for better harmonization techniques. The authors suggest domain-invariant machine learning as a potential future direction.
  • Prognosis Model Complexity: For predicting MCI to AD progression, the combined morphometry and connectome model did not significantly improve performance over morphometry alone. This might be due to the smaller sample size for this specific task, making feature selection for the connectome model suboptimal, or it may indicate a need for more sophisticated methods to combine multimodal data for prognosis.
  • Healthy Controls: The NHIS-IH cohort lacked a true healthy control group, using SMC individuals instead. While SMC individuals are cognitively normal, subtle differences in brain health might exist compared to truly healthy controls, potentially influencing classification. The ADNI-2 data, with its healthy control group, helped address this.
  • Future Directions: The authors suggest that their findings support the utility of individualized white matter connectomes, combined with machine learning, as a powerful tool for detecting AD-related neurodegeneration. Future work could focus on improving MRI data harmonization and developing more advanced ensemble methods for combining multimodal data for improved predictive power.

Summary updated

Abstract

ABSTRACT Accurate, reliable prediction of risk for Alzheimer’s disease (AD) is essential for early, disease-modifying therapeutics. Multimodal MRI, such as structural and diffusion MRI, is likely to contain complementary information of neurodegenerative processes in AD. Here we tested the utility of the multimodal MRI (T1-weighted structure and diffusion MRI), combined with high-throughput brain phenotyping—morphometry and structural connectomics—and machine learning, as a diagnostic tool for AD. We used, firstly, a clinical cohort at a dementia clinic (National Health Insurance Service-Ilsan Hospital [NHIS-IH]; N=211; 110 AD, 64 mild cognitive impairment [MCI], and 37 cognitively normal with subjective memory complaints [SMC]) to test the diagnostic models; and, secondly, Alzheimer’s Disease Neuroimaging Initiative (ADNI)-2 to test the generalizability. Our machine learning models trained on the morphometric and connectome estimates (number of features=34,646) showed optimal classification accuracy (AD/SMC: 97% accuracy, MCI/SMC: 83% accuracy; AD/MCI: 97% accuracy) in NHIS-IH cohort, outperforming a benchmark model (FLAIR-based white matter hyperintensity volumes). In ADNI-2 data, the combined connectome and morphometry model showed similar or superior accuracies (AD/HC: 96%; MCI/HC: 70%; AD/MCI: 75% accuracy) compared with the CSF biomarker model (t-tau, p-tau, and Amyloid β, and ratios). In predicting MCI to AD progression in a smaller cohort of ADNI-2 (n=60), the morphometry model showed similar performance with 69% accuracy compared with CSF biomarker model with 70% accuracy. Our comparison of classifiers trained on structural MRI, diffusion MRI, FLAIR, and CSF biomarkers show the promising utility of the white matter structural connectomes in classifying AD and MCI in addition to the widely used structural MRI-based morphometry, when combined with machine learning. Highlights We showed the utility of multimodal MRI, combining morphometry and white matter connectomes, to classify the diagnosis of AD and MCI using machine learning. In predicting the progression from MCI to AD, the morphometry model showed the best performance. Two independent clinical datasets were used in this study: one for model building, the other for generalizability testing.
BibTeX
@misc{wang2018diagnosis,
  author = {Yun Wang and Chenxiao Xu and Ji Hwan Park and Seonjoo Lee and Yaakov Stern and Shinjae Yoo and Jong Hun Kim and Hyoung Seop Kim and Jiook Cha and The Alzheimer’s Disease Neuroimaging Initiative},
  title = {{Diagnosis and Prognosis of Alzheimer’s Disease Using Brain Morphometry and White Matter Connectomes}},
  year = {2018},
  doi = {10.1101/407601},
  howpublished = {bioRxiv},
}
Figure