ADNI Dataset Research Articles

A fuzzy rule-based approach via MATLAB for the CDR instrument for staging the severity of dementia

Background: The CDR scale is a standard qualitative staging instrument that has been widely applied for assessing the severity of dementia which is based on information elicited through a semi-structured interview standardized in an assessment protocol. Despite clinical skills to elicit appropriate information are required, subjectivity still lies in the administration of the protocol and scoring process of the CDR. In this paper we propose a fuzzy rule-based CDR instrument to stage dementia based on the usual CDR, aiming to cover the subjectivities of the scoring process in the usual CDR which are directly related to the scoring system. This is effectively achieved by the F-CDR, our proposed expert system, which allows assigning scores continuously throughout the interval [0,3].Results: In order to test the performance of our fuzzy model, we compare the outputs FCDR obtained from of F-CDR approach to the outputs U-CDR obtained by a usual application of the CDR via the same inputs for both. The dataset provided by ADNI, composed of more than eleven thousand CDR tests, including the inputs and outputs (U-CDR), is the source for comparisons.Methods: The fuzzy rule-based model for the CDR that we propose in this paper is a fuzzy inference system (FIS) constructed in MATLAB with the aid of the Fuzzy Logic Designer app. The FIS was constructed based on the CDR and the specialist’s indications and tested on real data provided by ADNI.Conclusion: The high accuracy of matches between U-CDR and F-CDR via the same inputs over random samples selected from the ADNI dataset suggests that the fuzzy approach to the CDR instrument here proposed is suitable to extend the scoring process of the usual CDR since the fuzzy approach allows the possibility of scoring continuously in the interval [0,3].

Domain-Prior-Induced Structural MRI Adaptation for Clinical Progression Prediction of Subjective Cognitive Decline.

Growing evidence shows that subjective cognitive decline (SCD) among elderly individuals is the possible pre-clinical stage of Alzheimer's disease (AD). To prevent the potential disease conversion, it is critical to investigate biomarkers for SCD progression. Previous learning-based methods employ T1-weighted magnetic resonance imaging (MRI) data to aid the future progression prediction of SCD, but often fail to build reliable models due to the insufficient number of subjects and imbalanced sample classes. A few studies suggest building a model on a large-scale AD-related dataset and then applying it to another dataset for SCD progression via transfer learning. Unfortunately, they usually ignore significant data distribution gaps between different centers/domains. With the prior knowledge that SCD is at increased risk of underlying AD pathology, we propose a domain-prior-induced structural MRI adaptation (DSMA) method for SCD progression prediction by mitigating the distribution gap between SCD and AD groups. The proposed DSMA method consists of two parallel feature encoders for MRI feature learning in the labeled source domain and unlabeled target domain, an attention block to locate potential disease-associated brain regions, and a feature adaptation module based on maximum mean discrepancy (MMD) for cross-domain feature alignment. Experimental results on the public ADNI dataset and an SCD dataset demonstrate the superiority of our method over several state-of-the-arts.

Predicting diagnosis 4 years prior to Alzheimer’s disease incident

This study employed a deep learning longitudinal model, graph convolutional and recurrent neural network (graph-CNN-RNN), on a series of brain structural MRI scans for AD prognosis. It characterized whole-brain morphology via incorporating longitudinal cortical and subcortical morphology and defined a probabilistic risk for the prediction of AD as a function of age prior to clinical diagnosis. The graph-CNN-RNN model was trained on half of the Alzheimer’s Disease Neuroimaging Initiative dataset (ADNI, n = 1559) and validated on the other half of the ADNI dataset and the Open Access Series of Imaging Studies-3 (OASIS-3, n = 930). Our findings demonstrated that the graph-CNN-RNN can reliably and robustly diagnose AD at the accuracy rate of 85% and above across all the time points for both datasets. The graph-CNN-RNN predicted the AD conversion from 0 to 4 years before the AD onset at ∼80% of accuracy. The AD probabilistic risk was associated with clinical traits, cognition, and amyloid burden assessed using [18F]-Florbetapir (AV45) positron emission tomography (PET) across all the time points. The graph-CNN-RNN provided the quantitative trajectory of brain morphology from prognosis to overt stages of AD. Such a deep learning tool and the AD probabilistic risk have great potential in clinical applications for AD prognosis.

The effect of Alzheimer’s disease risk factors on brain aging in normal Chineses: Cognitive aging and cognitive reserve

Aging has been recognized as a major driving force of the Alzheimer’s disease’s (AD) progression, however, the relationship between brain aging and AD is still unclear. There is also a lack of studies investigating the influence of AD risk factors on brain aging in cognitively normal people. Here, the “Brain Age Gap Estimation” (BrainAGE) framework was applied to investigate the effects of AD risk factors on individual brain aging. Across a total of 165 cognitively normal elderly subjects, although no significant difference was observed in the BrainAGE scores among the three groups, AD risk dose (i.e., the number of AD risk factors) is tend to associated with an increased BrainAGE scores (high-risk > middle risk > low risk). Female exhibited more advanced brain aging (P = 0.004), and higher education years were associated with preserved brain aging (P < 0.001). APOE-ɛ4 (P = 0.846) and family history (FH) of dementia (P = 0.209) did not increase BrainAGE scores. When comparing 52 aMCI patients with 38 cognitively normal controls from ADNI dataset, aMCI patients showed significantly increased BrainAGE scores. BrainAGE scores were negatively correlated with CSF Aβ42 levels in the aMCI group (r = −0.275, P = 0.048). With an accuracy of 68.9%, BrainAGE outperformed APOE-ɛ4 and hippocampus gray matter volume (GMV) in predicting aMCI. In conclusion, AD is independently associated with structural changes in the brain that reflect advanced aging. Potentially, BrainAGE combined with APOE-ɛ4 and hippocampus GMV could be used as a pre-screening tool in early-stage AD.

RPOP: Robust PET-only processing of community acquired heterogeneous amyloid-PET data

The reference standard for amyloid-PET quantification requires structural MRI (sMRI) for preprocessing in both multi-site research studies and clinical trials. Here we describe rPOP (robust PET-Only Processing), a MATLAB-based MRI-free pipeline implementing non-linear warping and differential smoothing of amyloid-PET scans performed with any of the FDA-approved radiotracers (18F-florbetapir/FBP, 18F-florbetaben/FBB or 18F-flutemetamol/FLUTE). Each image undergoes spatial normalization based on weighted PET templates and data-driven differential smoothing, then allowing users to perform their quantification of choice. Prior to normalization, users can choose whether to automatically reset the origin of the image to the center of mass or proceed with the pipeline with the image as it is. We validate rPOP with n = 740 (514 FBP, 182 FBB, 44 FLUTE) amyloid-PET scans from the Imaging Dementia—Evidence for Amyloid Scanning – Brain Health Registry sub-study (IDEAS-BHR) and n = 1,518 scans from the Alzheimer's Disease Neuroimaging Initiative (n = 1,249 FBP, n = 269 FBB), including heterogeneous acquisition and reconstruction protocols. After running rPOP, a standard quantification to extract Standardized Uptake Value ratios and the respective Centiloids conversion was performed. rPOP-based amyloid status (using an independent pathology-based threshold of ≥24.4 Centiloid units) was compared with either local visual reads (IDEAS-BHR, n = 663 with complete valid data and reads available) or with amyloid status derived from an MRI-based PET processing pipeline (ADNI, thresholds of >20/>18 Centiloids for FBP/FBB). Finally, within the ADNI dataset, we tested the linear associations between rPOP- and MRI-based Centiloid values. rPOP achieved accurate warping for N = 2,233/2,258 (98.9%) in the first pass. Of the N = 25 warping failures, 24 were rescued with manual reorientation and origin reset prior to warping. We observed high concordance between rPOP-based amyloid status and both visual reads (IDEAS-BHR, Cohen's k = 0.72 [0.7–0.74], ∼86% concordance) or MRI-pipeline based amyloid status (ADNI, k = 0.88 [0.87–0.89], ∼94% concordance). rPOP- and MRI-pipeline based Centiloids were strongly linearly related (R2:0.95, p<0.001), with this association being significantly modulated by estimated PET resolution (β= -0.016, p<0.001). rPOP provides reliable MRI-free amyloid-PET warping and quantification, leveraging widely available software and only requiring an attenuation-corrected amyloid-PET image as input. The rPOP pipeline enables the comparison and merging of heterogeneous datasets and is publicly available at https://github.com/leoiacca/rPOP.

White matter hyperintensities are associated with grey matter atrophy and cognitive decline in Alzheimer's disease and frontotemporal dementia

White matter hyperintensities (WMHs) are commonly assumed to represent non-specific cerebrovascular disease comorbid to neurodegenerative processes, rather than playing a synergistic role. We compared the impact of WMHs on grey matter (GM) atrophy and cognition in normal aging (n = 571), mild cognitive impairment (MCI, n = 551), Alzheimer's dementia (AD, n = 212), fronto-temporal dementia (FTD, n = 125), and Parkinson's disease (PD, n = 271). Longitudinal data were obtained from ADNI, FTLDNI, and PPMI datasets. Mixed-effects models were used to compare WMHs and GM atrophy between patients and controls and assess the impact of WMHs on GM atrophy and cognition. MCI, AD, and FTD patients had significantly higher WMH loads than controls. WMHs were related to GM atrophy in insular and parieto-occipital regions in MCI/AD, and frontal regions and basal ganglia in FTD. In addition, WMHs contributed to more severe cognitive deficits in AD and FTD compared to controls, whereas their impact in MCI and PD was not significantly different from controls. These results suggest potential synergistic effects between WMHs and proteinopathies in the neurodegenerative process in MCI, AD and FTD.

A genetic exploration of cognitive decline in Alzheimer's disease.

The rate of cognitive decline in Alzheimer's disease (AD) has been found to vary widely between individuals, with numerous factors driving this heterogeneity. This study aims to explore the genetic architecture of cognitive decline in AD. We used a subset of 543 individuals with AD from the GERAD cohort (Harold et al., 2009) and a subset of 180 individuals with AD from the ADNI dataset (Mueller et al., 2005) with genetic data and cognitive assessments at multiple timepoints. Firstly, a metric of cognitive decline for each individual was estimated as the random slope from a random effects regression model of Mini Mental State Examination (MMSE) scores. This metric was consequently used to perform genome-wide association studies of cognitive decline in each dataset, which were then meta-analysed. Two polygenic risk score (PRS) analyses were performed, one exploring the polygenic heritability of decline in AD and the other examining the shared genetic architecture between decline and AD risk, based on the latest AD GWAS (Kunkle et al., 2019). Meta-analysis revealed 8 variants reaching genome-wide significance. Rate of decline PRS based on summary statistics from the GERAD data was not associated with decline in the ADNI dataset. Finally, AD risk PRS was not associated with rate of decline in either dataset. The lack of association of the AD risk PRS with rate of decline might indicate that the genetic risk for accelerated cognitive decline within AD is separate from the genetic risk for developing AD. The power of this study is limited by the small sample sizes, and replication in a larger dataset is necessary both to validate these findings and uncover additional variants that affect the rate of decline in AD.

RPOP: Robust PET‐only processing and quantification of community‐acquired amyloid‐PET from the IDEAS Study

AbstractBackgroundThe Imaging Dementia‐Evidence for Amyloid Scanning (IDEAS) Study collected over 15,000 community‐acquired amyloid‐PET scans, without structural MRI and with different acquisition times, tracers and scanners. Here we describe and validate rPOP (robust PET‐Only Processing), a pipeline for 18F‐Florbetapir (FBP), 18F‐Florbetaben (FBB) and 18F‐Flutemetamol (FLUTE) scans collected in IDEAS.MethodEach image undergoes automatic origin reset to center of mass, spatial normalization based on weighted PET templates (with SPM12) and data‐driven differential smoothing (with AFNI). We received n=740 (514 FBP, 182 FBB, 44 FLUTE) amyloid‐PET scans from the IDEAS–Brain Health Registry substudy (IDEAS‐BHR), with amyloid‐PET local visual reads available for N=663 scans. We performed semi‐quantification via rPOP and estimated neocortical Centiloids after tracer‐specific calibrations using the GAAIN VOIs with whole cerebellum reference. Centiloids were calculated for clinical groups (MCI or dementia), to compare rPOP‐based amyloid‐status (centiloids≥24.4) with visual reads. The pipeline was validated with n=1518 ADNI scans (n=1249 FBP, n=269 FBB), using MRI‐based amyloid‐status as a reference and testing the associations between rPOP‐ and MRI‐based centiloids.ResultrPOP successfully processed N=2233/2258 (98.9%) scans in the first pass, with N=24/25 warping‐related failures rescued with a manual reorientation and origin reset prior to warping. In IDEAS‐BHR, we observed the expected centiloids distribution (Figure 1A‐B) both according to visual reads and also stratifying by clinical stage amyloid‐negative, i.e. mean±sd Centiloids MCI/Dementia: 2±23/‐5±27; amyloid‐positive: MCI/Dementia: 65±42/80±37. There was high concordance (Table 1) between rPOP‐based amyloid status and both local visual reads (IDEAS‐BHR, Cohen’s k=0.72 [0.7‐0.74], ∼86% concordance). Very high concordance was also observed in the ADNI dataset (k=0.88 [0.87‐0.89], ∼94% concordance), with rPOP‐ and MRI‐based Centiloids being strongly linearly associated (R2:0.95, p&lt;0.001), with a tighter relationship in images with better resolution (β= ‐0.016, p&lt;0.001).ConclusionWith rPOP, we successfully derived Centiloids quantification from heterogeneous IDEAS amyloid‐PET scans and validated the pipeline with ADNI scans acquired in a research setting. These data show that community‐acquired amyloid‐PET scans can be successfully analyzed, enabling joint efforts by clinical sites and research centers.

Free water in T2 FLAIR white matter hyperintensity lesions

AbstractBackgroundWhite matter (WM) free water (FW) is likely associated with cerebral small vessel disease (CSVD). FW is the fraction of unconstrained water within an image voxel, which can be estimated from diffusion‐weighted images. T2‐weighted Fluid‐Attenuated Inversion Recovery (FLAIR) white matter hyperintensity (WMH) is a widely used index to assess the damages caused by CSVD. It is critical to characterize how FW content is altered in WMH lesions. In this work, we proposed a data processing framework to assess FW distributions in WMH and normal‐appearing WM as well as in different WM fiber tracts.MethodSingle‐shell diffusion‐weighted image (SS‐DWI) and T2 FLAIR image data of 133 cognitively normal (CN) adults (Table 1) were obtained from the ADNI dataset (http://adni.loni.usc.edu). FW maps were generated from SS‐DWI images using the DIPY software package (https://dipy.org/). WMH lesions were identified with the lesion segmentation toolbox implemented in SPM. We co‐registered FW and WMH maps of individual subjects in the MNI space. Then we derived group WM fiber tracts using the tract‐based spatial statistics (TBSS). Finally, we examined the localization of WMH and FW in the major WM tracts.ResultFigure 1 shows the FW map, FLAIR image, and WMH lesion from a representative subject. Figure 2 shows group average and individual FW distributions in WMH and normal‐appearing WM. Two sample T‐Test found that FW content in WMH is significantly higher than that in normal‐appearing WM (mean FW and standard deviation FW in WMH are 0.430 and 0.112, mean FW and standard deviation FW in normal‐appearing WM are 0.203 and 0.06, P &lt; 1×10‐30). TBSS based analysis showed that: 1) WMH lesions are mainly located in the corona radiata, posterior thalamic radiation, tapetum, superior fronto‐occipital fasciculus, superior longitudinal fasciculus, corpus callosum, and sagittal stratum; 2) Compared with mean FW in normal‐appearing WM, elevated FW content is mainly at the fornix, tapetum, cingulum hippocampus, superior fronto‐occipital fasciculus, corpus callosum, sagittal stratum, and posterior thalamic radiation.ConclusionWe developed an image processing pipeline to estimate FW changes in WMH lesions and normal‐appearing WM. Free water content is elevated in WMH lesions, which is likely driven by CSVD.

Effect of CSF biomarkers on cortical thickness in males and females

AbstractBackgroundThe level of elevated Tau, phosphorylated Tau, and aggregated Amyloid β in the cerebrospinal fluid (CSF) are markers of Alzheimer’s pathology in the brain. In this study, we analyzed how these CSF biomarkers are related to the cortical thickness in males and females, for subjects who are cognitively normal (CN), with mild cognitive impairment (MCI), and diagnosed as Alzheimer’s disease (AD).MethodWe studied the CN (305), MCI (264) and AD (230) subjects from the ADNI dataset at baseline. We first studied the CSF biomarker level differences between males and females in all three groups. The effect of age was controlled using general linear models (GLM) We also looked at the effect of the CSF biomarkers on cortical thickness maps generated separately for males and females for all three groups, also with GLM‐based covariate regression to control the effect of age, ICV, and field strength of the scan.ResultFor the CSF biomarker levels, we observed Tau to be significantly higher in females in the AD group. No other significant sex difference in the biomarker levels were observed. We found MCI males [Figure 1] and AD [Figure 2] females showing a negative relationship between Tau and thickness. When comparing CSF pTau levels to cortical thickness, a negative correlation was found only in males [Figure 3] for the AD group. When comparing Aβ and cortical thickness, negative correlations were found for males in the CN [Figure 4] and AD group [Figure 5], and for females [Figure 6] in the MCI group.ConclusionThe results suggest that the relationship between cortical thickness and CSF biomarkers is different in males and females in CN, MCI, and AD groups. We found that there is a negative relationship between CSF biomarkers and cortical thickness.

Anxiety throughout Alzheimer’s disease progression: In mice and (wo)men

Neuropsychiatric disturbances, such as depression and anxiety, are observed in 90% of Alzheimer's disease (AD) patients and are frequent in those at risk for AD (Lyketsos CG, et al. 2012). Similarly, epidemiological studies show that neurodegeneration and clinical symptoms occur more rapidly in females once diagnosed. In addition to cognitive decline and daily anxiety, AD patients (∼50%) become confused, agitated, and anxious in the evening, often referred to as "sundowning"(Canevelli M, et al. 2016). This circadian disruption can also greatly impact AD pathology. Restlessness before and during sleep has been shown to inhibit Ab clearance, increase neurofibrillary tangles, and dysregulate synaptic activity, all of which are characteristic hallmarks of AD (Cordone S, et al. 2019). Using our activity-dependent tagging system, the ArcCreERT2 x (ChR2)-EYFP x AD (APP/PS1), we labeled neurons activated during learning. These neurons can then can be compared with secondary neuronal ensembles activated during memory retrieval. The neurons activated at both time points represent an engram. Here, we aimed to identify the neural ensembles linking anxiety and memory loss following AD progression by utilizing behavioral studies, calcium imaging and whole-brain microscopy, in female and male mice. We found: 1) Female AD mice exhibited anxiety-like behavior at an earlier age compared to controls and male mice, 2) AD female mice displayed memory deficits as early as 2 months of age, 3) Anxiety-like behavior correlated with memory impairment only in AD female mice, 4) Unlike their male counterparts, female AD mice showed a decline in memory traces in the CA3 of the hippocampus, and 5) Male AD mice exhibit sundowning behavior at 2 months which correlates to increased calcium transients in the vCA1. We have also translated these findings to humans using the ADNI dataset. In humans, anxiety predicts transition to dementia and that anxiety has a sex-specific effect on brain atrophy. To use anxiety as a neuropsychiatric biomarker of AD in the human population in combination with current imaging and cognitive testing and to find novel brain areas associated with increased anxiety and memory loss. This could provide new therapeutic targets for those at risk for AD.

The normative data of cortical volumes in normal aging from Neuro I, a brain image quantitative analysis system

AbstractBackgroundThe aging of brain is a topic has been constantly studied. Particularly, the norms for the elderly are important in diagnosing degenerative brain diseases, but the aging brain of East Asians has not been fully studied yet. “Neuro I” is a brain image quantitative analysis system with a built‐in large‐scale normative database for East Asians to provide quantified clinical information on how different an individual patient is from normal people. The study aimed to investigate ethnic differences of cortical brain volumes in normal elderly people, and to determine whether the norms considering racial disparities could improve the accuracy of diagnosis.MethodUsing Neuro I, cortical volumes were measured from brain magnetic resonance (MR) images of 1,686 cognitively normal elderly Koreans aged 59 to 89. For Caucasians, the brain images of 855 Caucasian‐Non‐Hispanic subjects were selected from the ADNI and OASIS datasets. The regression model predicting brain volumes were built using ethnicity, age, sex, intracranial volume (ICV), magnetic field strength (MFS), and scanner manufacturers as predictors. Based on the prediction model, the z‐scores of each individual were calculated. To validate the z‐score for diagnosis, 420 Alzheimer’s disease (AD) subjects of Koreans and Caucasians, and the same number of normal subjects were selected. The four validation groups divided by race and diagnosis were age‐ and sex‐matched using a propensity score matching method.ResultThe models predicting cortical volumes had a considerate amount of explained variance (mean R2: 55.8%, range: 42.7%–77.5%). Ethnicity explained substantial variances for temporal and parietal lobes (mean R2: 5.6%, range: 4.2%–7.5%), comparable to age (mean R2: 3.3%, range: 2.0%–4.3%). Furthermore, the z score of each region based on the model with ethnicity as a predictor (AUCs = .75, .72, .67, and .67 for left and right temporal and left and right parietal lobes, respectively) improved the accuracy of AD diagnosis compared to the z score based on the model without ethnicity (AUCs = .62, .59, .55, and .57).ConclusionRace or ethnicity should be considered for establishing the normative data in normal aging and AD diagnosis.

Impact of polygenic risk score on normative models of hippocampal volumes

AbstractBackgroundNormative Models are widely used across clinical applications from infant growth charts to brain volume charts. For Alzheimer’s disease (AD), reduction in hippocampal volume (HV) is an early and often presymptomatic feature. Thus, precisely measuring deviation of HV from healthy aging facilitates earlier diagnosis. Brain volumes have been reported with heritability of 40‐90%, thus, implying that normative ranges of HV are heavily influenced by genetics. Here we investigate whether genetic information can improve HV nomograms for AD.MethodWe used Gaussian Process Regression (GPR) to generate nomograms of HV from 40,000 participants in UKBB aged 45‐82. These nomograms were contrasted to a previously reported Sliding Window method (SWM; PMID: 31254939). Next, we computed a Polygenic Risk Score (PRS) for HV using a large independent GWAS (N=33,536; PMID: 28098162). Participants were stratified into high and low PRS for HV; nomograms were built for each strata separately. Next, we used subjects from the ADNI dataset labelled cognitively normal (CN) or AD (N=226 and N=122) to examine the difference between using stratified versus non‐stratified nomograms.ResultNomograms generated using GPR matched those generated with the SWM with an average of difference of 17 mm3 (about 0.4%). Compared to the SWM, GPR extended the age range of the nomograms by ten years in either direction (from 56‐73 y/o to 47‐83 y/o: Figure 1). Nomograms stratified by PRS showed an average shift of 28 mm3 upward/downward for high/low PRS scoring samples (Figure 1). In ADNI, using non‐stratified nomograms led to a difference of 13% in median HV percentiles between high‐PRS and low‐PRS CN participants, which was reduced to 3.3% using stratified nomograms (Figure 2); this effect was not observed for the AD group.ConclusionGPR‐based nomograms were similar to the SWM‐nomograms while extending the range into areas critical for the analysis of aging diseases. Including genetics through a PRS for HV shifted HV nomograms in the expected direction. The small shift in nomograms of HV did not affect the sensitivity of AD diagnosis. However, it led to better adjusted estimates within CN subjects. Thus, genetics‐adjusted HV estimates may improve biomarker‐based AD‐risk predictions (PMID: 31526625).

FDN-ADNet: Fuzzy LS-TWSVM based deep learning network for prognosis of the Alzheimer’s disease using the sagittal plane of MRI scans

Alzheimer’s disease (AD) is the most pervasive form of dementia, resulting in severe psychosocial effects such as affecting personality, reasoning, emotions, and memory. Several neuroimaging techniques are available to correctly identify the structural changes in the brain, out of which the most popular is structural T-1 weighted Magnetic Resonance Imaging (MRI). From 3D MRI, sagittal plane slices provide more clear information related to the hippocampus, amygdala, corpus callosum, and several vital regions of the brain, which defines the extent of degeneration of the AD. Although diverse analysis of machine learning (ML) and deep learning (DL) based algorithm is already proposed for diagnosis of AD, still there is scope of research for early prediction so that treatment can be started either by medication or by improving the lifestyle. This paper proposed a DL model for all level feature extraction and fuzzy hyperplane based least square twin support vector machine (FLS-TWSVM) for the classification of the extracted features for early diagnosis of AD (FDN-ADNet) using extracted sagittal plane slices from 3D MRI images. Model is trained over the online available ADNI dataset and triangular fuzzy function is applied for the construction of hyperplane for classification. The proposed model attains the highest accuracy of 97.15%, 97.29% and 95% for CN vs AD, CN vs MCI and AD vs MCI classification, respectively when compared with the several state of the art networks.

Automatic Prediction of Cognitive and Functional Decline Can Significantly Decrease the Number of Subjects Required for Clinical Trials in Early Alzheimer's Disease.

Background:While both cognitive and magnetic resonance imaging (MRI) data has been used to predict progression in Alzheimer’s disease, heterogeneity between patients makes it challenging to predict the rate of cognitive and functional decline for individual subjects.Objective:To investigate prognostic power of MRI-based biomarkers of medial temporal lobe atrophy and macroscopic tissue change to predict cognitive decline in individual patients in clinical trials of early Alzheimer’s disease.Methods:Data used in this study included 312 patients with mild cognitive impairment from the ADNI dataset with baseline MRI, cerebrospinal fluid amyloid-β, cognitive test scores, and a minimum of two-year follow-up information available. We built a prognostic model using baseline cognitive scores and MRI-based features to determine which subjects remain stable and which functionally decline over 2 and 3-year follow-up periods.Results:Combining both sets of features yields 77%accuracy (81%sensitivity and 75%specificity) to predict cognitive decline at 2 years (74%accuracy at 3 years with 75%sensitivity and 73%specificity). When used to select trial participants, this tool yields a 3.8-fold decrease in the required sample size for a 2-year study (2.8-fold decrease for a 3-year study) for a hypothesized 25%treatment effect to reduce cognitive decline.Conclusion:When used in clinical trials for cohort enrichment, this tool could accelerate development of new treatments by significantly increasing statistical power to detect differences in cognitive decline between arms. In addition, detection of future decline can help clinicians improve patient management strategies that will slow or delay symptom progression.

Multiple Connection Pattern Combination From Single-Mode Data for Mild Cognitive Impairment Identification.

Mild cognitive impairment (MCI) is generally considered to be a key indicator for predicting the early progression of Alzheimer’s disease (AD). Currently, the brain connection (BC) estimated by fMRI data has been validated to be an effective diagnostic biomarker for MCI. Existing studies mainly focused on the single connection pattern for the neuro-disease diagnosis. Thus, such approaches are commonly insufficient to reveal the underlying changes between groups of MCI patients and normal controls (NCs), thereby limiting their performance. In this context, the information associated with multiple patterns (e.g., functional connectivity or effective connectivity) from single-mode data are considered for the MCI diagnosis. In this paper, we provide a novel multiple connection pattern combination (MCPC) approach to combine different patterns based on the kernel combination trick to identify MCI from NCs. In particular, sixty-three MCI cases and sixty-four NC cases from the ADNI dataset are conducted for the validation of the proposed MCPC method. The proposed method achieves 87.40% classification accuracy and significantly outperforms methods that use a single pattern.

A new unsupervised pseudo-siamese network with two filling strategies for image denoising and quality enhancement

Digital image noise may be introduced during acquisition, transmission, or processing and affects readability and image processing effectiveness. The accuracy of established image processing techniques, such as segmentation, recognition, and edge detection, is adversely impacted by noise. There exists an extensive body of work which focuses on circumventing such issues through digital image enhancement and noise reduction, but this work is limited by a number of constraints including the application of non-adaptive parameters, potential loss of edge detail information, and (with supervised approaches) a requirement for clean, labeled, training data. This paper, developed on the principle of Noise2Void, presents a new unsupervised learning approach incorporating a pseudo-siamese network. Our method enables image denoising without the need for clean images or paired noise images, instead requiring only noise images. Two independent branches of the network utilize different filling strategies, namely zero filling and adjacent pixel filling. Then, the network employs a loss function to improve the similarity of the results in the two branches. We also modify the Efficient Channel Attention module to extract more diverse features and improve performance on the basis of global average pooling. Experimental results show that compared with traditional methods, the pseudo-siamese network has a greater improvement on the ADNI dataset in terms of quantitative and qualitative evaluation. Our method therefore has practical utility in cases where clean images are difficult to obtain.

Diagnosis of Alzheimer's Disease with Ensemble Learning Classifier and 3D Convolutional Neural Network.

Alzheimer’s disease (AD), the most common type of dementia, is a progressive disease beginning with mild memory loss, possibly leading to loss of the ability to carry on a conversation and respond to environments. It can seriously affect a person’s ability to carry out daily activities. Therefore, early diagnosis of AD is conducive to better treatment and avoiding further deterioration of the disease. Magnetic resonance imaging (MRI) has become the main tool for humans to study brain tissues. It can clearly reflect the internal structure of a brain and plays an important role in the diagnosis of Alzheimer’s disease. MRI data is widely used for disease diagnosis. In this paper, based on MRI data, a method combining a 3D convolutional neural network and ensemble learning is proposed to improve the diagnosis accuracy. Then, a data denoising module is proposed to reduce boundary noise. The experimental results on ADNI dataset demonstrate that the model proposed in this paper improves the training speed of the neural network and achieves 95.2% accuracy in AD vs. NC (normal control) task and 77.8% accuracy in sMCI (stable mild cognitive impairment) vs. pMCI (progressive mild cognitive impairment) task in the diagnosis of Alzheimer’s disease.

DIAGNOSIS OF ALZHEIMER'S DISEASE BY THREE-DIMENSIONAL CONVOLUTIONAL NEURAL NETWORK USING UNSUPERVISED FEATURE LEARNING METHOD

The rise of Deep Learning in the past two decades has prompted research into solutions to help improve Alzheimer’s diagnosis based on neuroimaging data. As such, a wide variety of different techniques have been used, but a clear turn towards the use of Convolutional Neural Networks (CNN) has been observed in the last decade. To effectively predicate Alzheimer's Disease (AD), this paper proposed a two stage method. The first stage involves learning the best representation of the training data using an improved sparse autoencoder (SAE), an unsupervised neural network. The second stage involves using a 3D-Convolutional Neural Network (3D-CNN) to differentiate between the health status and diseased status based on the learned records and MRI scan of the brain. The SAE was optimized so as to train an efficient model. We report on experiments using the ADNI data set involving 897 historical scans. We demonstrate that using 3D convolutional neural networks with sparse auto encoder outperform several other classifiers stated in the literature.

A cascaded nested network for 3T brain MR image segmentation guided by 7T labeling

Accurate segmentation of the brain into gray matter, white matter, and cerebrospinal fluid using magnetic resonance (MR) imaging is critical for visualization and quantification of brain anatomy. Compared to 3T MR images, 7T MR images exhibit higher tissue contrast that is contributive to accurate tissue delineation for training segmentation models. In this paper, we propose a cascaded nested network (CaNes-Net) for segmentation of 3T brain MR images, trained by tissue labels delineated from the corresponding 7T images. We first train a nested network (Nes-Net) for a rough segmentation. The second Nes-Net uses tissue-specific geodesic distance maps as contextual information to refine the segmentation. This process is iterated to build CaNes-Net with a cascade of Nes-Net modules to gradually refine the segmentation. To alleviate the misalignment between 3T and corresponding 7T MR images, we incorporate a correlation coefficient map to allow well-aligned voxels to play a more important role in supervising the training process. We compared CaNes-Net with SPM and FSL tools, as well as four deep learning models on 18 adult subjects and the ADNI dataset. Our results indicate that CaNes-Net reduces segmentation errors caused by the misalignment and improves segmentation accuracy substantially over the competing methods.