Voxel-Based Statistical Analysis of 3D Immunostained Tissue Imaging

Michel E. Vandenberghe, Nicolas Souedet, Anne‐Sophie Hérard, Anne‐Marie Ayral, Florent Letronne, Yaël Balbastre, Elmahdi Sadouni, Philippe Hantraye, Marc Dhénain, Frédérique Frouin, Jean‐Charles Lambert, Thierry DelzescauxFrontiers in Neuroscience 2018 · DOI

TL;DR

This paper introduces a novel voxel-based statistical analysis method for 3D immunostained rodent brain images, enabling precise localization of pathological changes. By adapting established voxel-based techniques to high-resolution microscopy data, the authors demonstrate a more sensitive and spatially resolved approach to studying neurodegenerative diseases like Alzheimer's.

What Problem Does This Paper Address?

Traditionally, studying brain tissue at a microscopic level involves sectioning, which destroys the 3D structure. While recent advancements allow for 3D imaging of entire rodent brains, quantitatively analyzing these large, high-resolution datasets for subtle changes between experimental groups has remained a challenge. Existing region-based analyses can miss localized effects or be confounded by opposing changes within a region. This paper tackles the problem of applying voxel-based statistical analysis, commonly used in lower-resolution MRI, to the much higher resolution and complexity of 3D immunostained brain tissue. This is crucial for identifying precise locations of molecular or cellular alterations that might be missed by broader regional comparisons.

Methods

The core of this work is adapting voxel-based analysis to 3D immunostained microscopy data. Key steps include:

  • Parametric Map Generation: High-resolution 3D IHC images are first processed to create lower-resolution "parametric maps." This involves segmenting the marker of interest (e.g., Aβ plaques) and then smoothing these segmented images. The paper evaluates three methods for this: simple image binning, Voronoi tessellation with binning, and Gaussian kernel smoothing with binning. Gaussian kernel smoothing, optimized using a bootstrap error minimization approach, was found to be the most effective for generating smooth, representative marker load estimates.
  • Voxel-Based Statistical Analysis: Once parametric maps are generated, standard voxel-based statistical tests (specifically, a two-tailed Student's t-test) are applied to compare marker load between groups at each voxel.
  • Cluster Inference: To address the multiple comparisons problem inherent in voxel-wise analysis, the authors employ non-parametric cluster mass inference. This method identifies significant clusters of altered marker load by summing t-statistic values within contiguous regions and assessing their significance through permutations.
  • Anatomical Localization: Identified clusters are then localized within a standard anatomical atlas (MICe mouse brain atlas and Paxinos atlas) to provide precise anatomical context.

Datasets

  1. Simulation Dataset: A synthetic dataset was generated using a Boolean-Poisson model to simulate Aβ plaques of varying sizes and distributions. This dataset was crucial for validating the parametric map generation techniques and assessing the accuracy of the voxel-based analysis by comparing inferred clusters to known ground truth.
  2. ADAM30 Study Dataset: This dataset comprises 3D IHC images of Aβ plaque deposition in transgenic mouse models of Alzheimer's disease. Specifically, it includes brains from APPSwe,Ind-ADAM30 mice (expressing ADAM30) and APPSwe,Ind-Cre-ADAM30 mice (conditional ADAM30 expression), allowing the researchers to study the effect of ADAM30 on Aβ pathology.

Key Insights and Ideas

  • Bridging the Resolution Gap: The primary innovation is adapting voxel-based analysis, typically used for lower-resolution imaging (like MRI), to the ultra-high resolution of 3D immunostained microscopy. This unlocks the potential for precise, localized findings in microscopic brain imaging.
  • Parametric Maps as the Key: Creating smoothed, lower-resolution parametric maps from high-resolution IHC data is essential. This step effectively balances the need for detail with the statistical requirements of voxel-based analysis, mitigating issues of intrinsic variability at the microscopic level.
  • Bootstrap for Optimization: Using bootstrap error to select the optimal Gaussian kernel smoothing parameter is a robust, data-driven approach that avoids arbitrary choices and improves the reliability of the parametric maps.
  • Cluster Inference for Significance: Employing cluster mass inference is vital for controlling false positives in high-dimensional voxel-based analyses, especially with limited sample sizes common in animal studies.

Key Results

  • Simulation Validation: The Gaussian kernel smoothing method, optimized via bootstrap, demonstrated the best performance in simulation studies, accurately recovering cluster shapes and achieving high F1 scores for cluster detection.
  • ADAM30 Effect Localization: In the Alzheimer's mouse model, voxel-based analysis revealed that ADAM30 over-expression significantly reduced Aβ plaque load in specific sub-regions of the hippocampus (dentate gyrus) and cerebral cortex (cingulate cortex, retrosplenial agranular cortex, and lateral septal nucleus).
  • Region vs. Voxel: The study highlights how region-based analysis, while useful, could miss or underestimate the localized effects of ADAM30, particularly in the cerebral cortex where the effect was spatially restricted. Voxel-based analysis provided a more precise and sensitive detection of these localized changes.
  • Differential Effects: Further analysis within detected clusters showed that ADAM30 reduced Aβ plaque number in the cortex but reduced Aβ plaque size in the hippocampus, suggesting different mechanisms of action in these distinct brain regions.

Limitations and Future Work

  • Template Arbitrariness: The choice of a reference brain for anatomical normalization was somewhat arbitrary. Future work could focus on developing more objective methods for generating templates for 3D immunostained images.
  • Simulation Scope: The simulation study primarily focused on density variations between groups. Expanding simulations to include variations in object shape or size, or combinations thereof, would further validate the method.
  • Cluster Inference: The study explored one method of cluster inference. Investigating alternative approaches, such as Random Field Theory or adjusted cluster size methods, could offer further refinements.
  • Adaptive Smoothing: The paper suggests exploring adaptive Gaussian smoothing to preserve more detail in dense regions while smoothing sparser areas, potentially improving parametric map generation.
  • Broader Applicability: While demonstrated for Aβ plaques, the methodology is presented as applicable to other markers and imaging modalities (e.g., light-sheet microscopy, optical tomography), with the caveat that segmentation and normalization steps might need adaptation.

Summary updated

Abstract

Recently developed techniques to visualize immunostained tissues in 3D and in large samples have expanded the scope of microscopic investigations at the level of the whole brain. Here, we propose to adapt voxel-based statistical analysis to 3D high-resolution images of the immunostained rodent brain. The proposed approach was first validated with a simulation dataset with known cluster locations. Then, it was applied to characterize the effect of ADAM30, a gene involved in the metabolism of the amyloid precursor protein, in a mouse model of Alzheimer's disease. This work introduces voxel-based analysis of 3D immunostained microscopic brain images and, therefore, opens the door to localized whole-brain exploratory investigation of pathological markers and cellular alterations.
BibTeX
@inproceedings{vandenberghe2018voxel,
  author = {Michel E. Vandenberghe and Nicolas Souedet and Anne‐Sophie Hérard and Anne‐Marie Ayral and Florent Letronne and Yaël Balbastre and Elmahdi Sadouni and Philippe Hantraye and Marc Dhénain and Frédérique Frouin and Jean‐Charles Lambert and Thierry Delzescaux},
  title = {{Voxel-Based Statistical Analysis of 3D Immunostained Tissue Imaging}},
  booktitle = {Frontiers in Neuroscience},
  year = {2018},
  doi = {10.3389/fnins.2018.00754},
}
Figure