Study of the Influence of Age in 18F-FDG PET Images Using a Data-Driven Approach and Its Evaluation in Alzheimer’s Disease

Jiehui Jiang, Yiwu Sun, Hucheng Zhou, Shaoping Li, Zhemin Huang, Ping Wu, Kuangyu Shi, Chuantao Zuo, Alzheimer’s Disease Neuroimaging InitiativeContrast 2018 · DOI

TL;DR

This study introduces a data-driven method to identify brain regions whose glucose metabolism (measured by 18F-FDG PET) declines with age. They developed an age-correction model that significantly improves the ability to detect Alzheimer's disease (AD) by reducing age-related noise in PET scans.

What Problem Does This Paper Address?

Diagnosing neurodegenerative diseases like Alzheimer's disease (AD) using brain imaging, particularly 18F-FDG PET scans, is challenging because normal aging itself causes changes in brain metabolism. Radiologists struggle to distinguish between age-related metabolic decline and disease-specific changes. This ambiguity can lead to misdiagnosis or delayed diagnosis. The paper aims to address this by developing a method to mathematically correct for the influence of age on 18F-FDG PET images, thereby enhancing the sensitivity for detecting AD-related abnormalities.

Methods

The researchers employed a data-driven approach using Independent Component Analysis (ICA), specifically FLICA (Functional MRI of the Brain's Linked Independent Component Analysis), to analyze brain metabolism patterns in healthy individuals. ICA is a technique that decomposes complex signals into underlying independent components, assuming these components represent distinct biological processes.

  1. Component Identification: FLICA was used to identify brain components that showed a strong negative correlation with age. This means finding brain regions where metabolic activity decreases as age increases.
  2. Region Definition: After identifying a key component (IC3) strongly linked to aging, they refined the specific brain regions within this component by applying various thresholds to voxel values. This process aimed to pinpoint the most sensitive areas for age-related metabolic changes.
  3. Age-Correction Model Development: A linear regression model was developed to quantify the relationship between age and the standardized uptake value ratio (SUVR) in the identified brain regions. This model allowed for the calculation of an "age effect" that could be subtracted from the original SUVR values.
  4. Validation: The developed model and identified brain regions were validated on independent datasets, including both healthy individuals and AD patients. This involved statistical analyses to assess the improvement in correlation coefficients and diagnostic accuracy after age correction.

Datasets

The study utilized two primary datasets:

  1. Huashan Hospital Dataset: 255 healthy control 18F-FDG PET images from Huashan Hospital, Fudan University, Shanghai, China. This dataset was used for the initial data-driven exploration and model development.
  2. ADNI Dataset: 262 healthy control and 50 Alzheimer's Disease (AD) 18F-FDG PET images selected from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. This dataset was used for validation of the identified regions and the age-correction model.

Key Insights and Ideas

  • Age as a Major Confounder: The paper strongly reinforces that age is a significant interfering factor in interpreting 18F-FDG PET scans for dementia diagnosis.
  • Specific Brain Regions Affected: The study identified a specific set of brain regions, including parts of the frontal, insula, cingulate, and temporal cortices, that exhibit a consistent metabolic decline with normal aging.
  • Data-Driven Approach for Discovery: Using ICA (FLICA) proved effective in uncovering these age-related patterns without prior assumptions about which regions would be affected.
  • Mathematical Correction is Feasible: The development of a linear age-correction model demonstrates that the influence of age on PET signal can be quantified and removed.
  • Improved Diagnostic Power: Age correction not only reduces age-related variability in healthy individuals but also significantly enhances the ability to differentiate between healthy aging and AD pathology.

Key Results

  • Identified Brain Regions: The study identified several brain regions, including the inferior frontal gyrus, superior frontal gyrus (medial and orbital parts), right insula, left anterior cingulate, left median cingulate, paracingulate gyri, and bilateral superior temporal gyri, as showing a strong negative correlation with age.
  • Strong Age-Metabolism Correlation: The key component (IC3) identified explained 51.8% of the variance in age-related metabolic changes.
  • Effective Age Correction: After applying the age-correction model to healthy subjects, the correlation between age and SUVR mean in the identified regions decreased significantly (e.g., from R = -0.212 to R = -0.027).
  • Enhanced AD Detection: For AD patients, age correction led to a significant improvement in the correlation between SUVR mean and clinical measures (MMSE, CDRSB, FDG). For instance, the correlation with MMSE improved from R = 0.371 to R = 0.497.
  • Increased Classification Accuracy: Using PCA and SVM for AD classification, accuracy increased from 79% before age correction to 90% after age correction. The Area Under the Curve (AUC) for ROC curves also improved from 0.735 to 0.84.

Limitations and Future Work

  • Limited Sample Size for Age-Specific Models: The study used a single age-correction model for the entire 20-80 year age span. Future work could develop age-specific models for finer precision.
  • Linear Model Assumption: The age-correction model was linear. While IC3 showed a strong linear trend, other components might exhibit non-linear age-related changes, suggesting the need for nonlinear models.
  • Lack of Partial Volume Correction (PVC): The absence of MRI data prevented PVC, which could have further refined the PET signal accuracy.
  • Geographical Differences: The study used datasets from China and the US (ADNI). Future research could investigate how geographical differences in populations might affect age-related brain patterns and the need for different templates.
  • Multimodal Analysis: The study focused solely on FDG PET. Integrating other imaging modalities (MRI, PiB-PET, DTI) could provide a more comprehensive understanding of age-related brain changes and lead to more robust correction methods.
  • Sex Differences: The paper notes that differences in natural brain decline between males and females need further study.

Summary updated

Abstract

All correlation coefficients were significantly improved after the age correction. The proposed model was effective in the age correction of both healthy and AD subjects.
BibTeX
@inproceedings{jiang2018study,
  author = {Jiehui Jiang and Yiwu Sun and Hucheng Zhou and Shaoping Li and Zhemin Huang and Ping Wu and Kuangyu Shi and Chuantao Zuo and Alzheimer’s Disease Neuroimaging Initiative},
  title = {{Study of the Influence of Age in <sup>18</sup>F-FDG PET Images Using a Data-Driven Approach and Its Evaluation in Alzheimer’s Disease}},
  booktitle = {Contrast},
  year = {2018},
  doi = {10.1155/2018/3786083},
}
Figure