Search

INTRODUCTION: Amnestic mild cognitive impairment (aMCI) is emerging as a heterogeneous condition. METHODS: We looked at a cohort of N = 207 aMCI subjects, with baseline fluorodeoxyglucose positron emission tomography (FDG‐PET), T1 magnetic resonance imaging, cerebrospinal fluid (CSF), apolipoprotein E (APOE), and neuropsychological assessment. An algorithm based on FDG‐PET hypometabolism classified each subject into subtypes, then compared biomarker measures and clinical progression. RESULTS: Three subtypes emerged: hippocampal sparing–cortical hypometabolism, associated with younger age and the highest level of Alzheimer's disease (AD)‐CSF pathology; hippocampal/cortical hypometabolism, associated with a high percentage of APOE ε3/ε4 or ε4/ε4 carriers; medial–temporal hypometabolism, characterized by older age, the lowest AD‐CSF pathology, the most severe hippocampal atrophy, and a benign course. Within the whole cohort, the severity of temporo‐parietal hypometabolism, correlated with AD‐CSF pathology and marked the rate of progression of cognitive decline. DISCUSSION: FDG‐PET can distinguish clinically comparable aMCI at single‐subject level with different risk of progression to AD dementia or stability. The obtained results can be useful for the optimization of pharmacological trials and automated‐classification models. Highlights: Algorithm based on FDG‐PET hypometabolism demonstrates distinct subtypes across aMCI;Three different subtypes show heterogeneous biological profiles and risk of progression;The cortical hypometabolism is associated with AD pathology and cognitive decline;MTL hypometabolism is associated with the lowest conversion rate and CSF‐AD pathology. [ABSTRACT FROM AUTHOR]

Background: Neurodegenerative diseases lead to memory disorders and cognitive impairments with major impacts on day‐to‐day function. Early detection of patients is essential for an efficient care approach, and differential diagnosis is essential for determining prognosis and tailoring management. Most research studies compare patients by pairwise comparison methods, but this does not correspond to daily clinical practice, in which clinicians need to distinguish many different possible types of dementia concomitantly. It has been shown that magnetic resonance imaging (MRI) can mark with good accuracy the different types of dementia, such as: Alzheimer's disease (AD), Frontotemporal dementia (FTD), Parkinson's disease (PD), dementia with Lewy bodies (DLB). Recently, Deep Learning approaches have started being used although differential diagnosis is still almost ignored. The aim of this study was to develop a Convolution Neural Network (CNN) to classify probable AD, FTD, DLB, PD and healthy elderly control (CN) subjects starting from Diffusion Tensor Imaging (DTI). Method: The cohort used in this study came from six different initiatives: ADNI1, NIFD2, NACC3, PDBP4, PPMI5 and Newcastle University6,7,8,9. We analysed a total of 108 CN, 110 AD, 135 FTD, 150 DLB, 132 PD. Demographic and neuropsychological characterization is shown in Table 1. The DTI scans were pre‐processed using the FMRIB software library (FSL) with eddy current correction, skull‐stripping and diffusion tensor calculation. We post‐processed the Fractional Anisotropy (FA) map using 3D CNN. The proposed CNN structure consisted of an image input layer, convolutional filters, several batch normalization with max‐pooling layers and, finally, the classification layer. A data augmentation approach was adopted that ultimately resulted in a dataset of 26'450 brain scans. A hold‐out validation (70% training set 30% test set) was performed. Result: We found excellent performance of the 3D CNN in discriminating the different neurodegenerative diseases. Figure 1 shows the confusion matrix with the following values for the area under the ROC curve in each class: 0.92 for CN, 0.90 for AD, 0.97 for DLB, 0.96 for FTD, 0.99 for PD. Conclusion: These results prove that Deep Learning and the 3D CCN method are suitable for clinical practice and provide additional clues that can help clinicians in diagnosis making. [ABSTRACT FROM AUTHOR]

Background: The choroid plexus (CP) is a structure located inside the brain ventricles and is responsible for the production of cerebrospinal fluid (CSF), transport of proteins across the epithelium to the CSF and clearance of proteins from the CSF. The CP undergoes changes with aging, which are exacerbated in Alzheimer's disease (AD) and could alter CP functioning. Nonetheless, the relation between CP functioning and AD pathophysiology is still unclear. Here, we investigated the association of CP volume with levels of CSF proteins predominantly expressed by the CP and amyloid positivity in non‐demented individuals. Method: We included 235 individuals with normal cognition (NC) or mild cognitive impairment (MCI) from the European EMIF‐AD MBD study (49% women, mean age 67.9 (SD 7.9) years). Based on CSF Aβ1‐42 (A, data‐driven cut‐offs) and p‐tau (T, center‐specific cut‐offs), individuals were classified as cognitively normal with normal A and T (controls, n=54). Based on CSF Aβ1‐42, individuals with NC or MCI were classified as amyloid‐positive (CN A+, n=61; MCI A+, n=67). CSF proteomic data were centrally generated using TMT mass spectrometry and 5 CSF proteins with predominant expression in the CP were selected based on the Protein Atlas, Harmonizome and literature (TTR, MFRP, SLC5A5, FOLR1, SOD3). CP and hippocampal volumes on MRI were quantified using FreeSurfer. We used non‐parametric (Spearman) partial correlations corrected for age, gender and total intracranial volume (TIV). Result: Lower Aβ42 levels were associated with higher CP volume in MCI (Figure 1). Higher levels of CP predominant expressed proteins were generally associated with increased CP volume in persons with amyloid positivity, in both preclinical and prodromal AD (Table 1). Lower Aβ42 levels were associated with lower hippocampal volume in MCI (Figure 2). Lower hippocampal volumes were associated with higher CP volumes in MCI (Figure 3). Nonetheless, hippocampal volume and the levels of CP predominant expressed proteins were generally not correlated (Table 2). Conclusion: Our findings suggest that CP volume is associated with levels of proteins predominantly expressed by the CP in predementia stages of AD. This would suggest CP dysfunction in early AD. This is important for future research and AD treatment development. [ABSTRACT FROM AUTHOR]

Background: The advent of big data and artificial intelligence has energized the neuroimaging community with powerful processing tools and the need for large‐scale computational resources for data analysis has increased. Not all neuroscientists have access to extensive resources or have sufficient expertise to configure these libraries, making them difficult to use. Method: A collection of popular MRI scan processing software, (i.e.: Fsl, FreeSurfer, Ants, SPM, Adaboost, Neuromorphometrics, LPA, Tracula) was selected as a proof of concept to be included in the NeuGRID Workflow to produce an advanced tool for group analysis and biomarkers assessment. Result: The NeuGRID Workflow was designed to allow the user to build a custom MRI processing pipeline by means of a layered structure (Fig.1) available for users at the URL https://neugrid2.eu/index.php/workflowgroup. In the first layer, the user must upload a compressed file containing T13D, FLAIR, and DWI scans of subjects. In the second layer, the user selects which kind of MRI sequence to process. If DICOMs were uploaded, the third layer provides tools to convert DICOMs into NIFTI, the default format for all subsequent workflow processes. All DICOMs are anonymized upon upload. The fourth layer allows the user to choose how images are preprocessed, and the preprocessing pipelines can be customized for different modalities. In the final layer, the user can choose how to analyze the images to calculate volume, thickness, area and curvature of brain regions, quantify white‐matter lesions and reconstruct major white‐matter pathways. The triggered job is efficiently parallelized using a Sun Grid Engine scheduler to reduce computation time. After the scans are processed in the NeuGRID cluster (450 CPUs, 2.5TB memory, 400TB storage), the user is notified by email that the job is finished and the results can be downloaded. Conclusion: The NeuGRID Workflow is a practical tool that allows researchers without computational resources and/or expertise to design a custom pipeline to process large amounts of data from anywhere using a traditional laptop or personal computer. The collection of heterogeneous tools provides a high degree of freedom in pipeline design, making NeuGRID Workflow a Swiss Army knife for MRI processing [ABSTRACT FROM AUTHOR]

Background: In recent years, statistical and machine learning algorithms allowed advancements in many fields of healthcare research, but such methods are still mostly used by technical experts. Their implementation in user‐friendly software/web interfaces is fundamental for their broader adoption by patient‐facing physicians. Method: A validated SuStaIn (Young et al, 2019, https://doi.org/10.1038/s41467-018-05892-0) model for atrophy patterns in Alzheimer's Disease (AD) was used as core for a web application. The model was trained using volumes of relevant brain regions of 1031 subjects from the Alzheimer's Disease Neuroimaging Initiative and validated using 731 subjects from the OASIS, PharmaCog and ViTA Cohorts. Biomarkers used to train and validate the model were corrected against the effects of age, sex, education, APOE genotype and total intracranial volume (TIV). The model was exploited in the web application for single‐case scenarios. The user is required to input age and sex of the subject and upload an MRI scan which is processed with Freesurfer 5.3 to extract relevant brain volumes. Volumes are corrected against the effects of sex, age and TIV and converted into z‐scores used to compute the most likely among the three subtypes defined by the model and stage the subject on the relative z‐scores progression sequence. Results are summarized in a report (Fig. 1) and sent to the user via e‐mail. The web application skips the correction against education and APOE genotype as these factors may be unavailable and their influence on volumes should be minor. The validity of the different correction procedure was tested by subtyping and staging 50 subjects from the training set using the web application. Outputs were compared with the results of the validated model. Result: The web application was built and deployed on the neuGRID platform, and it is available at https://www.neugrid2.eu/index.php/ml-tools/sustain/. All subjects tested on the web application were assigned the same subtype as the validated model, although an average difference equal to 2.2 positions was observed for the staging. Conclusion: This web application implementing a SuStaIn model for AD atrophy subtypes represents a first example of advanced data‐driven algorithms implemented in a user‐friendly tool usable by clinicians in daily practice. [ABSTRACT FROM AUTHOR]

Background Hippocampal volumetry on magnetic resonance imaging is recognized as an Alzheimer's disease (AD) biomarker, and manual segmentation is the gold standard for measurement. However, a standard procedure is lacking. We operationalize and quantitate landmark differences to help a Delphi panel converge on a set of landmarks. Methods One hundred percent of anatomic landmark variability across 12 different protocols for manual segmentation was reduced into four segmentation units (the minimum hippocampus, the alveus/fimbria, the tail, and the subiculum), which were segmented on magnetic resonance images by expert raters to estimate reliability and AD-related atrophy. Results Intra- and interrater reliability were more than 0.96 and 0.92, respectively, except for the alveus/fimbria, which were 0.86 and 0.77, respectively. Of all AD-related atrophy, the minimum hippocampus contributed to 67%; tail, 24%; alveus/fimbria, 4%; and the subiculum, 5%. Conclusions Anatomic landmark variability in available protocols can be reduced to four discrete and measurable segmentation units. Their quantitative assessment will help a Delphi panel to define a set of landmarks for a harmonized protocol. [ABSTRACT FROM AUTHOR]

Background This study aimed to have international experts converge on a harmonized definition of whole hippocampus boundaries and segmentation procedures, to define standard operating procedures for magnetic resonance (MR)-based manual hippocampal segmentation. Methods The panel received a questionnaire regarding whole hippocampus boundaries and segmentation procedures. Quantitative information was supplied to allow evidence-based answers. A recursive and anonymous Delphi procedure was used to achieve convergence. Significance of agreement among panelists was assessed by exact probability on Fisher's and binomial tests. Results Agreement was significant on the inclusion of alveus/fimbria ( P = .021), whole hippocampal tail ( P = .013), medial border of the body according to visible morphology ( P = .0006), and on this combined set of features ( P = .001). This definition captures 100% of hippocampal tissue, 100% of Alzheimer’s disease-related atrophy, and demonstrated good reliability on preliminary intrarater (0.98) and inter-rater (0.94) estimates. Discussion Consensus was achieved among international experts with respect to hippocampal segmentation using MR resulting in a harmonized segmentation protocol. [ABSTRACT FROM AUTHOR]

Background The European Alzheimer's Disease Consortium and Alzheimer's Disease Neuroimaging Initiative (ADNI) Harmonized Protocol (HarP) is a Delphi definition of manual hippocampal segmentation from magnetic resonance imaging (MRI) that can be used as the standard of truth to train new tracers, and to validate automated segmentation algorithms. Training requires large and representative data sets of segmented hippocampi. This work aims to produce a set of HarP labels for the proper training and certification of tracers and algorithms. Methods Sixty-eight 1.5 T and 67 3 T volumetric structural ADNI scans from different subjects, balanced by age, medial temporal atrophy, and scanner manufacturer, were segmented by five qualified HarP tracers whose absolute interrater intraclass correlation coefficients were 0.953 and 0.975 (left and right). Labels were validated as HarP compliant through centralized quality check and correction. Results Hippocampal volumes (mm 3 ) were as follows: controls: left = 3060 (standard deviation [SD], 502), right = 3120 (SD, 897); mild cognitive impairment (MCI): left = 2596 (SD, 447), right = 2686 (SD, 473); and Alzheimer’s disease (AD): left = 2301 (SD, 492), right = 2445 (SD, 525). Volumes significantly correlated with atrophy severity at Scheltens' scale (Spearman's ρ = <−0.468, P = <.0005). Cerebrospinal fluid spaces (mm 3 ) were as follows: controls: left = 23 (32), right = 25 (25); MCI: left = 15 (13), right = 22 (16); and AD: left = 11 (13), right = 20 (25). Five subjects (3.7%) presented with unusual anatomy. Conclusions This work provides reference hippocampal labels for the training and certification of automated segmentation algorithms. The publicly released labels will allow the widespread implementation of the standard segmentation protocol. [ABSTRACT FROM AUTHOR]

Background: Most of the studies that focus on the Alzheimer's disease (AD) prediction forecast a dementia label: cognitively normal, mild cognitive impairments or AD. However these labels cannot always be compared across cohorts, are not perfectly accurate and describe the disease thanks to a limited number of stages (only 3) which, in practice, is a continuous progression. For these reasons, we predict a cognitive test (MMSE) that has a finer discretization and is similarly assessed across cohorts. Method: The Bayesian mixed‐effects model introduced in (Schiratti et al. 2015, NIPS) allows to estimate the average temporal progression of a subset of scores from ADNI, out of the individual longitudinal data. This profile is personalized to individuals from the ADNI cohort but also AIBL and PharmaCOG. It allows to predict future values (1.5 to 6 years after the last seen visit). To measure the error of prediction, the value to predict is hidden and then compared to the prediction. This error is compared to the (i) constant prediction (i.e. no MMSE change), (ii) linear and (iii) logit regressions and (iv) the test‐retest MMSE noise ‐ whose standard deviation is 2.8 out of 30 (Clark et al, 1999, Archives of Neurology) ‐ which is the best achievable score. Result: The constant prediction, while being trivial, is better than the linear and logit regressions [Figure 1]. Moreover, it is of the order of the noise (i.e. best possible prediction) for horizons of less than 3 years. In any case, our prediction is of the noise order or smaller than the constant prediction, allowing to predict accurately the MMSE at any temporal horizon. Moreover, the accuracy on the AIBL and PharmaCOG [Figures 2 and 3] datasets highlights the potential of the model to generalize to patients from other cohorts but also in real‐life applications. Conclusion: The constant prediction outperforms standard regression techniques (linear and logit) due to the important noise level of the MMSE and the fact that the progression of the disease occurs over long periods of time. However, our model is able to beat this prediction for far enough temporal horizons. [ABSTRACT FROM AUTHOR]

Background: Understanding the path that leads from mild cognitive impairment (MCI) to Alzheimer's disease (AD) provides insights into dementia pathophysiology and can inform patient stratification in clinical trials. Our objective is to train statistical computational models for estimating an average disease trajectory to better understand how patients evolve on the basis of biomarkers of from the ADNI population, then validate the model on MCI subjects from the PharmaCog cohort. Method: The disease trajectory was built via a Gaussian Process Progression Model (GPPM) (Lorenzi et al. 2017, DOI: 10.1016/j.neuroimage.2017.08.059) trained on longitudinal biomarker data from 338 subjects in the ADNI database who had converted from MCI to AD at the time of writing. Biomarkers related to cognitive scores, cerebrospinal fluid and T13D magnetic resonance imaging were used to build the disease model. Validation was performed by staging 139 MCI subjects from the PharmaCog database, 20 of whom progressed to AD, along the progression model timeline. Only baseline cross sectional measures of biomarkers were used to stage PharmaCog subjects in order to simulate clinical settings. Measures of sensitivity, specificity, balanced accuracy and area under curve (AUC) were used to measure the staging performance. Result: The staging of PharmaCog subjects shows a clear separation between MCI stable (sMCI) subjects and MCI progressors (pMCI) on the disease timeline. On average, sMCIs were staged at year 72.7±3.6 while pMCIs are staged 3 years later (p‐value<0.01) at year 75.5±2.8. Classification of sMCI vs pMCI subjects returned a sensitivity equal to 0.68, specificity equal to 0.85, balanced accuracy equal to 0.77 and the ROC curve (Figure 1) had AUC equal to 0.75. Conclusion: We identified MCI converters to AD in a clinical data set (PharmaCog) using a data‐driven computational model of AD progression trained on research data (ADNI). Classification performance was comparable to other data‐driven tools (Young et al., 2014, DOI: 10.1093/brain/awu176), but performed in a more challenging clinical setting. This validates the staging efficacy of such models and shows their utility for future application in healthcare and clinical trials. [ABSTRACT FROM AUTHOR]

Background: Heterogeneity of cohorts, in terms of inclusion criteria, design of follow‐up visits and batteries of cognitive assessments, hinders any thorough comparisons between them. For that reason, we build a cross‐cohort model of cognitive decline that can be personalized to any patient, allowing to impute partially or totally missing scores. This enables to compare the subjects' progression across cohorts thanks to common cognitive assessments. Method: Based on a generic framework of disease progression that handles longitudinal data (Schiratti et al. 2015, NIPS) and implemented in the Leaspy Python package, we first estimate the joint progression of a large set of cognitive scores from the ADNI database along a common disease timeline. This group‐average progression is then personalized to individual trajectories of patients from the PharmaCog and AIBL cohorts. In particular, all individuals are characterized by a relative pace of progression and onset age that can be compared across cohorts. To validate our model, we impute missing values of a given score, based on its past and future values together with the dynamics of concurrent scores and compare errors to constant predictions and linear regressions. We finally go further by simulating purposely concealed scores thanks to the progression of the remaining features so to mock reconstruction of scores that are entirely missing in a particular cohort. Results: Figure 1 shows the relative positioning of all subjects from the 3 cohorts based on their individual pace of progression and their age at disease onset. Such characterization enables to impute missing values of the MMSE or the Immediate/Delayed Logical Memory better than imputing the last known value or fitting a linear regression [Figure 2]. Furthermore, from the multiple measurements of a single score, the model is able to reconstruct the individual progression of the other scores at the corresponding visits [Figure 3], with a controlled error rate. Conclusion: Our model enables comparison of patients trajectories from different cohorts. It permits to lessen imputation errors of most cognitive scores in new cohorts and demonstrates a new way to simulate, at any individual visit, scores that have not been measured in practice. [ABSTRACT FROM AUTHOR]

Background: The abundance of patient data collected during clinical routine is rarely harnessed towards progress in neuroscience. However, multimodal real‐life data may provide significant insights into pathophysiology, disease progression and prognosis, in particular in the field of Alzheimer's disease. The aim of the Medical Informatics Platform (MIP) within the Human Brain Project is to create an intuitive interface that allows remote computational analyses of clinical data stored in the respective medical centers. Method: The concept of federated analyses refers to the statistical processing of physically separate and non‐aggregated data sets in a single user interface. The MIP interface provides advanced statistical approaches, remote data access and local user regulations. Authorized MIP users are able to test their hypotheses using a flexible portfolio of data sets, modalities and data processing tools. Result: At the present stage, three memory clinics have contributed to the MIP: the Leenaards Memory Centre at the University Hospital Lausanne, Switzerland; the University Hospital Lille Memory Centre, France; and the IRCCS Istituto Centro San Giovanni Di Dio Fatebenefratelli di Brescia, Italy. The data sets contain demographic, neuropsychological and volumetric MRI data from 5154 individuals with subjective and mild cognitive impairment or dementia. The data include clinically probable Alzheimer's etiology according to CSF biomarkers. Proof‐of‐concept and validation analyses will be presented. Reference data from the Alzheimer's Disease Neuroimaging Initiative (ADNI), which can be downloaded by MIP users in accordance with ADNI rules, can be compared with the clinical data in the MIP. Conclusion: Federated analyses such as afforded by the MIP within the Human Brain Project do not only allow novel scientific insights due to a substantial increase in sample size and cross‐validation, but also ensure a high level of data protection. The MIP has a considerable potential for discovery in clinical neuroscience. [ABSTRACT FROM AUTHOR]