Background: Aβ pathology is associated with longitudinal changes of brain metabolism, atrophy, and cognition, in cognitively healthy elders. However, Aβ information is usually measured cross-sectionally and dichotomized to classify subjects as Aβ-positive or Aβ-negative, making it difficult to evaluate when brain and cognitive changes occur with respect to emerging Aβ pathology. In this study, we use longitudinal Aβ information to combine the level and rate of change of Aβ to estimate the time to Aβ-positivity for each subject and test this temporal proximity to significant Aβ pathology for associations with brain structure, metabolism, and cognition. Methods: In 89 cognitively healthy elders with up to 10 years of follow-up, we estimated the points at which rates of fluorodeoxyglucose (FDG) PET, MRI, and cognitive and functional decline begin to accelerate with respect to the time to Aβ-positivity. Points of initial acceleration in rates of decline were estimated using mixed-effects models with penalized regression splines. Results: Acceleration of rates of FDG PET were observed to occur 20+ years before the conventional threshold for Aβ-positivity. Subtle signs of cognitive dysfunction were observed 10+ years before Aβ-positivity. Conclusions: Aβ may have subtle associations with other hallmarks of Alzheimer's disease before Aβ biomarkers reach conventional thresholds for Aβ-positivity. Therefore, we propose that emerging Aβ pathology occurs many years before cognitively healthy elders reach the current threshold for Aβ positivity (preclinical AD). To allow prevention in the earliest disease stages, AD clinical trials may be designed to also include subjects with Aβ biomarkers in the sub-threshold range.

Cortical thinning patterns in Alzheimer's disease (AD) have been widely reported through conventional regional analysis. In addition, the coordinated variance of cortical thickness in different brain regions has been investigated both at the individual and group network levels. In this study, we aim to investigate network architectural characteristics of a structural covariance network (SCN) in AD, and further to show that the structural covariance connectivity becomes disorganized across the brain regions in AD, while the normal control (NC) subjects maintain more clustered and consistent coordination in cortical atrophy variations. We generated SCNs directly from T1-weighted MR images of individual patients using surface-based cortical thickness data, with structural connectivity defined as similarity in cortical thickness within different brain regions. Individual SCNs were constructed using morphometric data from the Samsung Medical Center (SMC) dataset. The structural covariance connectivity showed higher clustering than randomly generated networks, as well as similar minimum path lengths, indicating that the SCNs are "small world." There were significant difference between NC and AD group in characteristic path lengths (z = -2.97, p < 0.01) and small-worldness values (z = 4.05, p < 0.01). Clustering coefficients in AD was smaller than that of NC but there was no significant difference (z = 1.81, not significant). We further observed that the AD patients had significantly disrupted structural connectivity. We also show that the coordinated variance of cortical thickness is distributed more randomly from one region to other regions in AD patients when compared to NC subjects. Our proposed SCN may provide surface-based measures for understanding interaction between two brain regions with co-atrophy of the cerebral cortex due to normal aging or AD. We applied our method to the AD Neuroimaging Initiative (ADNI) data to show consistency in results with the SMC dataset.

Methods: Alzheimer's disease and Frontotemporal dementia are the first and third most common forms of dementia. Due to their similar clinical symptoms, they are easily misdiagnosed as each other even with sophisticated clinical guidelines. For disease-specific intervention and treatment, it is essential to develop a computer-aided system to improve the accuracy of their differential diagnosis. Recent advances in deep learning have delivered some of the best performance for medical image recognition tasks. However, its application to the differential diagnosis of AD and FTD pathology has not been explored. Approach: In this study, we proposed a novel deep learning based framework to distinguish between brain images of normal aging individuals and subjects with AD and FTD. Specifically, we combined the multi-scale and multi-type MRI-base image features with Generative Adversarial Network data augmentation technique to improve the differential diagnosis accuracy. Results: Each of the multi-scale, multitype, and data augmentation methods improved the ability for differential diagnosis for both AD and FTD. A 10-fold cross validation experiment performed on a large sample of 1,954 images using the proposed framework achieved a high overall accuracy of 88.28%. Conclusions: The salient contributions of this study are three-fold: (1) our experiments demonstrate that the combination of multiple structural features extracted at different scales with our proposed deep neural network yields superior performance than individual features; (2) we show that the use of Generative Adversarial Network for data augmentation could further improve the discriminant ability of the network regarding challenging tasks such as differentiating dementia sub-types; (3) and finally, we show that ensemble classifier strategy could make the network more robust and stable.

Determination of sensitive and specific markers of very early AD progression is intended to aid researchers and clinicians to develop new treatments and monitor their effectiveness, as well as to lessen the time and cost of clinical trials. Magnetic Resonance (MR)-related biomarkers have been recently identified by the use of machine learning methods for the in vivo differential diagnosis of AD. However, the vast majority of neuroimaging papers investigating this topic are focused on the difference between AD and patients with mild cognitive impairment (MCI), not considering the impact of MCI patients who will (MCIc) or not convert (MCInc) to AD. Morphological T1-weighted MRIs of 137 AD, 76 MCIc, 134 MCInc, and 162 healthy controls (CN) selected from the Alzheimer's disease neuroimaging initiative (ADNI) cohort, were used by an optimized machine learning algorithm. Voxels influencing the classification between these AD-related pre-clinical phases involved hippocampus, entorhinal cortex, basal ganglia, gyrus rectus, precuneus, and cerebellum, all critical regions known to be strongly involved in the pathophysiological mechanisms of AD. Classification accuracy was 76% AD vs. CN, 72% MCIc vs. CN, 66% MCIc vs. MCInc (nested 20-fold cross validation). Our data encourage the application of computer-based diagnosis in clinical practice of AD opening new prospective in the early management of AD patients.

Objective: It is well-known that the hippocampus presents significant asymmetry in Alzheimer's disease (AD) and that difference in volumes between left and right exists and varies with disease progression. However, few works investigated whether the asymmetry degree of subfields of hippocampus changes through the continuum from Mild Cognitive Impairment (MCI) to AD. Thus, aim of the present work was to evaluate the Asymmetry Index (AI) of hippocampal substructures as possible MRI biomarkers of Dementia. Moreover, we aimed to assess whether the subfields presented peculiar differences between left and right hemispheres. We also investigated the relationship between the asymmetry magnitude in hippocampal subfields and the decline of verbal memory as assessed by Rey's auditory verbal learning test (RAVLT). Methods: Four-hundred subjects were selected from ADNI, equally divided into healthy controls (HC), AD, stable MCI (sMCI), and progressive MCI (pMCI). The structural baseline T1s were processed with FreeSurfer 6.0 and volumes of whole hippocampus (WH) and 12 subfields were extracted. The AI was calculated as: (|Left-Right|/(Left+Right))*100. ANCOVA was used for evaluating AI differences between diagnoses, while paired t-test was applied for assessing changes between left and right volumes, separately for each group. Partial correlation was performed for exploring relationship between RAVLT summary scores (Immediate, Learning, Forgetting, Percent Forgetting) and hippocampal substructures AI. The statistical threshold was Bonferroni corrected p < 0.05/13 = 0.0038. Results: We found a general trend of increased degree of asymmetry with increasing severity of diagnosis. Indeed, AD presented the higher magnitude of asymmetry compared with HC, sMCI and pMCI, in the WH (AI mean 5.13 ± 4.29 SD) and in each of its twelve subfields. Moreover, we found in AD a significant negative correlation (r = -0.33, p = 0.00065) between the AI of parasubiculum (mean 12.70 ± 9.59 SD) and the RAVLT Learning score (mean 1.70 ± 1.62 SD). Conclusions: Our findings showed that hippocampal subfields AI varies differently among the four groups HC, sMCI, pMCI, and AD. Moreover, we found-for the first time-that hippocampal substructures had different sub-patterns of lateralization compared with the whole hippocampus. Importantly, the severity in learning rate was correlated with pathological high degree of asymmetry in parasubiculum of AD patients.

Alzheimer's disease (AD) is the most common neurodegenerative disorder worldwide and is one of the leading sources of morbidity and mortality in the aging population. There is a long preclinical period followed by mild cognitive impairment (MCI). Clinical diagnosis and the rate of decline is variable. Progression monitoring remains a challenge in AD, and it is imperative to create better tools to quantify this progression. Brain magnetic resonance imaging (MRI) is commonly used for patient assessment. However, current approaches for analysis require strong a priori assumptions about regions of interest used and complex preprocessing pipelines including computationally expensive non-linear registrations and iterative surface deformations. These preprocessing steps are composed of many stacked processing layers. Any error or bias in an upstream layer will be propagated throughout the pipeline. Failures or biases in the non-linear subject registration and the subjective choice of atlases of specific regions are common in medical neuroimaging analysis and may hinder the translation of many approaches to the clinical practice. Here we propose a data-driven method based on an extension of a deep learning architecture, DeepSymNet, that identifies longitudinal changes without relying on prior brain regions of interest, an atlas, or non-linear registration steps. Our approach is trained end-to-end and learns how a patient's brain structure dynamically changes between two-time points directly from the raw voxels. We compare our approach with Freesurfer longitudinal pipelines and voxel-based methods using the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. Our model can identify AD progression with comparable results to existing Freesurfer longitudinal pipelines without the need of predefined regions of interest, non-rigid registration algorithms, or iterative surface deformation at a fraction of the processing time. When compared to other voxel-based methods which share some of the same benefits, our model showed a statistically significant performance improvement. Additionally, we show that our model can differentiate between healthy subjects and patients with MCI. The model's decision was investigated using the epsilon layer-wise propagation algorithm. We found that the predictions were driven by the pallidum, putamen, and the superior temporal gyrus. Our novel longitudinal based, deep learning approach has the potential to diagnose patients earlier and enable new computational tools to monitor neurodegeneration in clinical practice.

Collecting multiple modalities of neuroimaging data on the same subject is increasingly becoming the norm in clinical practice and research. Fusing multiple modalities to find related patterns is a challenge in neuroimaging analysis. Canonical correlation analysis (CCA) is commonly used as a symmetric data fusion technique to find related patterns among multiple modalities. In CCA-based data fusion, principal component analysis (PCA) is frequently applied as a preprocessing step to reduce data dimension followed by CCA on dimension-reduced data. PCA, however, does not differentiate between informative voxels from non-informative voxels in the dimension reduction step. Sparse PCA (sPCA) extends traditional PCA by adding sparse regularization that assigns zero weights to non-informative voxels. In this study, sPCA is incorporated into CCA-based fusion analysis and applied on neuroimaging data. A cross-validation method is developed and validated to optimize the parameters in sPCA. Different simulations are carried out to evaluate the improvement by introducing sparsity constraint to PCA. Four fusion methods including sPCA+CCA, PCA+CCA, parallel ICA and sparse CCA were applied on structural and functional magnetic resonance imaging data of mild cognitive impairment subjects and normal controls. Our results indicate that sPCA significantly can reduce the impact of non-informative voxels and lead to improved statistical power in uncovering disease-related patterns by a fusion analysis.