Gebre, Robel K

AbstractBackgroundThe dynamics of white matter (WM) changes are understudied in Alzheimer’s disease (AD). Leveraging tau deposition measured from Tau‐PET, our goal was to study the association between tau and white matter health using NODDI from multi‐shell diffusion MRI (dMRI). Specifically, we focused on the Neurite Density Index (NDI) from NODDI, which in WM captures the signal from axons.MethodWe identified 251 participants (126 cognitively normal, 58 MCI, 67 dementia) from the Mayo ADRC and Mayo Clinic Study of Aging with Aβ‐PET, Tau‐PET, and multi‐shell dMRI who were Aβ‐positive (mean age of 77 years and 55% male). We limited to Aβ‐positive participants because Aβ deposition accelerates tau deposition. For Tau‐PET, we estimated SUVRs from three regions corresponding to Braak I/II, III/IV, and V/VI (Figure 1.a). To understand the associations between Tau‐PET SUVRs at three Braak regions and NDIs at 30 candidate WM tracts, we implemented three analyses methods ‐ Pearson correlation, linear regression, and multilayer perceptron machine learning regression.ResultHigher tau burden in all three Braak regions was associated with lower neurite density in 10 significant WM tracts (Table 1). Interestingly, uncinate fasciculus (UNC) was found significantly associated with all three Braak regions. All three analysis methods consistently showed differential involvement of WM pathways for Braak staging such as ‐ Cingulum adjoining Hippocampus (CGH) for Braak I/II regions, Medial Temporal WM (MTWM) for Braak III/IV regions, and Inferior Occipital WM (IOWM) for Braak V/VI regions (Figure 1.b). The linear regression and machine learning regression models also suggested that WM measurements reflected 18‐35% of variance due to tau depositionConclusionWe found that CGH and UNC WM were the most impacted ROIs associated with Braak I/II regions as expected. UNC, directly connecting anterior temporal lobes to the orbitofrontal cortex, was a key WM tract identified as important for all Braak I‐VI regions and warrants further investigation in the study of AD progression. This study highlights the significant impact tau deposition may have on WM. This work also demonstrates the feasibility of using NODDI as an informative supplementary imaging modality for monitoring AD progression.

AbstractBackgroundNeurite Orientation Dispersion and Density Imaging (NODDI) is becoming increasingly common as multiband acceleration becomes available for clinical scanning, but fitting the model can be extremely slow. Since the release of the original MATLAB implementation in 2012, three alternatives have become popular: Accelerated Microstructure Imaging via Convex Optimization (AMICO), the Microstructure Diffusion Toolbox (MDT), and the CUDA Diffusion Modelling Toolbox (cuDIMOT). They accelerate fitting through reformulation and/or using GPUs, but the implementation differences raise the question of which approach is best. To our knowledge all previous comparisons lacked knowledge of the true values and used the original implementation as the nominally correct one, with in vivo data. We remedy that using simulations with known NODDI values, in addition to in vivo scans.MethodWe created a software phantom with Neurite Density Index (NDI) values linearly increasing from 0.1 to 1.0 in x, Orientation Dispersion (ODI) from 0 to 1 in y, and Isotropic water Volume Fraction (ISOVF) from 0 to 0.9 in z. We simulated five scans using the ADNI3 advanced diffusion MRI protocol. The phantom had 20 voxels in each direction, and the fiber bundle principal direction randomly varied between the simulations. The simulations were fit by the original, AMICO, MDT, and cuDIMOT implementations, and root mean square errors were calculated as shown in Table 1. A discrepancy measure was also calculated using the brain voxels of three in vivo human scans acquired using the same protocol on a Siemens Prisma, substituting the original implementation’s values for the truth since the latter is unknown for in vivo scans.ResultcuDIMOT had the least error (for ODI < 0.8) and was fastest. The original implementation was a close second in accuracy but much slower. cuDIMOT’s ODI underestimation can be corrected with a transfer function in the noise‐free case, but that approach fails with actual scans.ConclusionThe original or cuDIMOT implementations should be used, with cuDIMOT the favorite when at least one CUDA‐capable GPU is available and the number of scans are large. Correcting cuDIMOT’s ODI underestimation using a different initialization or form of Dawson’s integral should be investigated.

AbstractBackgroundCerebral Microbleeds (CMBs) are tracked in anti‐amyloid drug safety monitoring. T2*GRE or SWI MRI images are visually assessed for the presence of CMBs. Automated methods tuned for sensitivity often have unacceptably high false positive rates requiring extensive human cleanup.MethodAssume contemporaneous T2*GRE and T1‐weighted images. CMB detection (Figures 1,2) has three steps: 1) Segmentation outputs from the T1‐weighted image are resampled into the T2*GRE image voxel raster; 2) Normalized cross‐correlation of the T2*GRE with a CMB template identifies “candidate CMBs”; 3) a neural network using voxel patches centered on each candidate CMB rejects false positives. Patches presented to the network are 4D and include five contrasts drawn from the T2*GRE, correlation, T1‐weighted, WM probability and CSF probability images. The simple network architecture consists of convolutional blocks followed by dense blocks. The system is trained using images from ADNI, ARIC, MCSA, and the Mayo Clinic ADRC that have been previously visually graded for CMBs (Table in Figure 3). Importantly, we include participants with and without CMBs in our data reflecting the fact that the majority of people (even the elderly) have no CMBs. Data is split 70:10:20 by subject for training, validation, and testing. Various network configurations were evaluated.ResultHigh candidate detection sensitivity can be achieved at the expense of presenting hundreds to thousands of false positive candidates per input image to the rejection network. Our best performing network configuration to date has AUC 0.982 (95% bootstrap CI 0.971 to 0.990 based on the 10% validation subset). Choosing an operating point with 95% sensitivity in the validation set and admits, on average, 0.6 false positive CMB candidates per input T2*GRE image. Sensitivity is hindered by a small number of obvious errors in the ground truth visual assessment. Validation loss tracks training loss with no indication of over‐fitting. Example patches are shown in Figure 4.ConclusionExcessive false positive detection requires significant human correction and effort, preventing automated CMB detection from being widely adopted. We have an automated detection system that is sensitive (95%) and provides excellent false positive rejection (<1 false positive CMB/input image) reducing needed human effort.

AbstractBackgroundData heterogeneity due to cross‐scanner variations is a major challenge in multi‐center aging and dementia studies. Brain structural measures for the same participant can vary due to scanner differences (manufacturer, scanner age/technology, signal‐to‐noise ratio, pulse sequence design, contrast, resolution) and result in biomarker variability. In this work, we applied basic image processing and deep learning methods (style transfer (ST) and generative adversarial networks (GANs)) for harmonizing T1‐weighted MRI scans across change in vendors (GE to Siemens) and resolution (1x1x1.2mm to 0.8x0.8x0.8mm) and test whether this harmonization improved compatibility of AD biomarker measures.MethodWe utilized training (n=814) and test (n=113 same participants scanned on two scanners close‐in‐time, typically on the same day, with different resolution) data from cognitively impaired and unimpaired participants scanned on GE and Siemens scanners from Mayo Clinic Study of Aging and Mayo ADRC. The harmonization was conducted to translate GE to Siemens. All scans were affine transformed to the same template space, resampled to 1.5 mm3, and bias corrected. Histogram matching was performed using cumulative density functions. A ResNet18 was trained for the ST’s content and style layers. Two types of GANs, conditional GANs (cGAN) and cycle‐consistent GANs (cycleGAN), were constructed. We then measured regional cortical thickness from each image using FreeSurfer‐7.1.1 and compared the thickness measurements from the original and translated scans using intraclass‐correlation (ICC) and Deming regression.ResultIn the test dataset, ST and cycleGAN showed the best performances when comparing the translated GE to Siemens. With ST, the ICC increased (vs. the original scans) slightly for the temporal meta‐ROI (0.956 from 0.954) and more substantially for the frontal pole ROI (0.706 from 0.594). The cycleGAN showed an increase in ICC for the frontal pole ROI (0.610 from 0.594) but not the meta‐ROI (Fig. 1).ConclusionEach method leveraged a different strength. Histogram matching captures contrast, ST extracts textural information, and GANs learn combination of contrast, edge and texture properties. We found that deep learning is a promising approach to cross‐scanner variability. Our comparisons revealed anatomical differences across methods; thus, disease specific measure should be considered for selection of the scanner harmonization method.

AbstractBackgroundCerebral Microbleeds (CMBs) are tracked in anti‐amyloid drug safety monitoring. T2*GRE or SWI MRI images are visually assessed for the presence of CMBs. Automated methods tuned for sensitivity often have unacceptably high false positive rates requiring extensive human cleanup.MethodAssume contemporaneous T2*GRE and T1‐weighted images. CMB detection (Figures 1,2) has three steps: 1) Segmentation outputs from the T1‐weighted image are resampled into the T2*GRE image voxel raster; 2) Normalized cross‐correlation of the T2*GRE with a CMB template identifies “candidate CMBs”; 3) a neural network using voxel patches centered on each candidate CMB rejects false positives. Patches presented to the network are 4D and include five contrasts drawn from the T2*GRE, correlation, T1‐weighted, WM probability and CSF probability images. The simple network architecture consists of convolutional blocks followed by dense blocks. The system is trained using images from ADNI, ARIC, MCSA, and the Mayo Clinic ADRC that have been previously visually graded for CMBs (Table in Figure 3). Importantly, we include participants with and without CMBs in our data reflecting the fact that the majority of people (even the elderly) have no CMBs. Data is split 70:10:20 by subject for training, validation, and testing. Various network configurations were evaluated.ResultHigh candidate detection sensitivity can be achieved at the expense of presenting hundreds to thousands of false positive candidates per input image to the rejection network. Our best performing network configuration to date has AUC 0.982 (95% bootstrap CI 0.971 to 0.990 based on the 10% validation subset). Choosing an operating point with 95% sensitivity in the validation set and admits, on average, 0.6 false positive CMB candidates per input T2*GRE image. Sensitivity is hindered by a small number of obvious errors in the ground truth visual assessment. Validation loss tracks training loss with no indication of over‐fitting. Example patches are shown in Figure 4.ConclusionExcessive false positive detection requires significant human correction and effort, preventing automated CMB detection from being widely adopted. We have an automated detection system that is sensitive (95%) and provides excellent false positive rejection (<1 false positive CMB/input image) reducing needed human effort.

AbstractBackgroundNeurite Orientation Dispersion and Density Imaging (NODDI) is becoming increasingly common as multiband acceleration becomes available for clinical scanning, but fitting the model can be extremely slow. Since the release of the original MATLAB implementation in 2012, three alternatives have become popular: Accelerated Microstructure Imaging via Convex Optimization (AMICO), the Microstructure Diffusion Toolbox (MDT), and the CUDA Diffusion Modelling Toolbox (cuDIMOT). They accelerate fitting through reformulation and/or using GPUs, but the implementation differences raise the question of which approach is best. To our knowledge all previous comparisons lacked knowledge of the true values and used the original implementation as the nominally correct one, with in vivo data. We remedy that using simulations with known NODDI values, in addition to in vivo scans.MethodWe created a software phantom with Neurite Density Index (NDI) values linearly increasing from 0.1 to 1.0 in x, Orientation Dispersion (ODI) from 0 to 1 in y, and Isotropic water Volume Fraction (ISOVF) from 0 to 0.9 in z. We simulated five scans using the ADNI3 advanced diffusion MRI protocol. The phantom had 20 voxels in each direction, and the fiber bundle principal direction randomly varied between the simulations. The simulations were fit by the original, AMICO, MDT, and cuDIMOT implementations, and root mean square errors were calculated as shown in Table 1. A discrepancy measure was also calculated using the brain voxels of three in vivo human scans acquired using the same protocol on a Siemens Prisma, substituting the original implementation’s values for the truth since the latter is unknown for in vivo scans.ResultcuDIMOT had the least error (for ODI < 0.8) and was fastest. The original implementation was a close second in accuracy but much slower. cuDIMOT’s ODI underestimation can be corrected with a transfer function in the noise‐free case, but that approach fails with actual scans.ConclusionThe original or cuDIMOT implementations should be used, with cuDIMOT the favorite when at least one CUDA‐capable GPU is available and the number of scans are large. Correcting cuDIMOT’s ODI underestimation using a different initialization or form of Dawson’s integral should be investigated.