Classification Of Cognitive Impairment Research Articles

Multiscale Analysis of Alzheimer's Disease Using Feature Fusion in Cognitive and Sensory Brain Regions

Introduction: This research is focused on early detection of Alzheimer's disease (AD) using a multiscale feature fusion framework, combining biomarkers from memory, vision, and speech regions extracted from Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET) images. Methods: Using 2D Gray Level Co-occurrence Matrix (2D-GLCM) texture features, volume, Standardized Uptake Value Ratios (SUVR), and obesity from different neuroimaging modalities, the study applies various classifiers, demonstrating a feature importance analysis in each Region of Interest (ROI). The research employs four classifiers, namely Linear Support Vector Machine (L-SVM), Linear Discriminant Analysis (LDA), Logistic Regression (LR), and Logistic Regression with Stochastic Gradient Descent (LRSGD) Classifiers, to determine feature importance, leading to subsequent validation using a Probabilistic Neural Network (PNN) Classifier. Results: The research highlights the critical role of brain texture features, particularly in memory regions, for AD detection. Significant sex-specific differences are observed, with males showing significance in texture features in memory regions, volume in vision regions, and SUVR in speech regions, while females exhibit significance in texture features in memory and speech regions, and SUVR in vision regions. Additionally, the study analyzes how obesity affects features used in AD prediction models, clarifying its effects on speech and vision regions, particularly brain volume. Conclusion: The findings contribute valuable insights into the effectiveness of feature fusion, sex-specific differences, and the impact of obesity on AD-related biomarkers, paving the way for future research in early AD detection strategies and cognitive impairment classification.

Cellular Automata Framework for Classification of Cognitive Impairment from Clinical Dementia Rating

Cellular Automata Framework for Classification of Cognitive Impairment from Clinical Dementia Rating

Longitudinal Patterns and Predictors of Cognitive Impairment Classification Stability.

Classifications such as Cognitive Impairment, No Dementia (CIND) are thought to represent the transitory, pre-clinical phase of dementia. However, increasing research demonstrates that CIND represents a nonlinear, unstable entity that does not always lead to imminent dementia. The present study utilizes a longitudinal repeated measures design to gain a thorough understanding of CIND classification stability patterns and identify predictors of future stability. The objectives were to i) explore patterns of longitudinal stability in cognitive status across multiple assessments and ii) investigate whether select baseline variables could predict 6-year CIND stability patterns. Participants (N = 259) included older adults (aged 65-90years) from Project MIND, a six-year longitudinal repeated measures design in which participants were classified as either normal cognition (NC) or CIND at each annual assessment. A latent transition analysis approach was adapted in order to identify and characterize transitions in CIND status across annual assessments. Participants were classified as either Stable NC, Stable CIND, Progressers, Reverters, or Fluctuaters. Multinomial logistic regression was employed to test whether baseline predictors were associated with cognitive status stability patterns. The sample demonstrated high rates of reversion and fluctuation in CIND status across annual assessments. Additionally, premorbid IQ and CIND severity (i.e., single vs. multi-domain impairment) at baseline were significantly associated with select stability outcomes. CIND status was unstable for several years following baseline assessment and cognitive reserve may delay or protect against demonstrable cognitive impairment. Further, consideration of cognitive impairment severity at the time of initial classification may improve CIND classifications.

Self- and study partner-reported cognitive decline in older adults without dementia: The role of α-synuclein and amyloid biomarkers in the Alzheimer's Disease Neuroimaging Initiative.

Subjective cognitive decline (SCD) may be an early marker of Alzheimer's disease (AD) pathology. Until recently, it was impossible to measure biomarkers specific for α-synuclein pathology; therefore, its association with subjective reports of cognitive decline is unknown. Alzheimer's Disease Neuroimaging Initiative participants without dementia (n=918) were classified as positive or negative for amyloid beta (Aβ+ or Aβ-) and α-synuclein (α-syn+ or α-syn-) biomarkers. Self- and study partner-reported cognitive decline was measured with the Everyday Cognition (ECog) questionnaire. Per self-report, Aβ+/α-syn+ had the greatest cognitive decline. Aβ-/α-syn+ did not differ from Aβ-/α-syn- across ECog scores. Study partner-reported results had a similar pattern, but Aβ+/α-syn- and Aβ+/α-syn+ did not differ across ECog scores. Mild cognitive impairment classification moderated the study partner-reported memory score. While α-syn+ alone did not increase subjective reports of cognitive decline, Aβ+/α-syn+ had the most self- and study partner-rated cognitive decline. Therefore, the presence of multiple pathologies was associated with greater SCD. Cerebrospinal fluid α-synuclein (α-syn) seed amplification assay was used to determine α-syn positivity. Amyloid beta (Aβ)-/α-syn-, Aβ-/α-syn+, Aβ+/α-syn-, and Aβ+/α-syn+ biomarker groups were created. Aβ+/α-syn+ had greater subjective cognitive decline (SCD) than the other biomarker groups. Aβ-/α-syn+ did not differ from Aβ-/α-syn- across self- or study-partner reported SCD scores. Study partner-reported subjective memory results were largely driven by participants with mild cognitive impairment.

Advancements in deep learning for Alzheimer's disease diagnosis: A comprehensive exploration and critical analysis of neuroimaging approaches

AbstractAlzheimer's disease (AD) is a major global health concern that affects millions of people globally. This study investigates the technical challenges in AD analysis and provides a thorough analysis of AD, emphasizing the disease's worldwide effects as well as the predicted increase. It explores the technological difficulties associated with AD analysis, concentrating on the shift in automated clinical diagnosis using MRI data from conventional machine learning to deep learning techniques. This study advances our knowledge of the effects of AD and provides new developments in deep learning for precise diagnosis, providing insightful information for both clinical and future research. The research introduces an innovative deep learning model, leveraging YOLOv5 and variants of YOLOv8, to classify AD images into four (NC, EMCI, LMCI, AD) categories. This study evaluates the performance of YOLOv5 which achieved high accuracy (97%) in multi‐class classification (classes 0 to 3) with precision, recall, and F1‐score reported for each class. YOLOv8 (Small) and YOLOv8 (Medium) models are also assessed for Alzheimer's disease diagnosis, demonstrating accuracy of 97% and 98%, respectively. Precision, recall, and F1‐score metrics provide detailed insights into the models' effectiveness across different classes. Comparative analysis against a transfer learning model reveals YOLOv5, YOLOv8 (Small), and YOLOv8 (Medium) consistently outperforming across six binary classifications related to cognitive impairment. These models show improved sensitivity and accuracy compared to baseline architectures from [32]. In AD/NC classification, YOLOv8 (Medium) achieves 98.43% accuracy and 97.45% sensitivity, for EMCI/LMCI classification, YOLOv8 (Medium) also excels with 92.12% accuracy and 90.12% sensitivity. The results highlight the effectiveness of YOLOv5 and YOLOv8 variants in neuroimaging tasks, showcasing their potential in clinical applications for cognitive impairment classification. The proposed models showcase superior performance, achieving high accuracy, sensitivity, and F1‐scores, surpassing baseline architectures and previous methods. Comparative analyses highlight the robustness and effectiveness of the proposed models in AD classification tasks, providing valuable insights for future research and clinical applications.

A Transformer Approach for Cognitive Impairment Classification and Prediction.

Early classification and prediction of Alzheimer disease (AD) and amnestic mild cognitive impairment (aMCI) with noninvasive approaches is a long-standing challenge. This challenge is further exacerbated by the sparsity of data needed for modeling. Deep learning methods offer a novel method to help address these challenging multiclass classification and prediction problems. We analyzed 3 target feature-sets from the National Alzheimer Coordinating Center (NACC) dataset: (1) neuropsychological (cognitive) data; (2) patient health history data; and (3) the combination of both sets. We used a masked Transformer-encoder without further feature selection to classify the samples on cognitive status (no cognitive impairment, aMCI, AD)-dynamically ignoring unavailable features. We then fine-tuned the model to predict the participants' future diagnosis in 1 to 3 years. We analyzed the sensitivity of the model to input features via Feature Permutation Importance. We demonstrated (1) the masked Transformer-encoder was able to perform prediction with sparse input data; (2) high multiclass current cognitive status classification accuracy (87% control, 79% aMCI, 89% AD); (3) acceptable results for 1- to 3-year multiclass future cognitive status prediction (83% control, 77% aMCI, 91% AD). The flexibility of our methods in handling inconsistent data provides a new venue for the analysis of cognitive status data.

Classification of Cognitive Impairment Using Quadratic Discriminant Analysis Based Spiral Dynamic Optimization Algorithm

Classification of Cognitive Impairment Using Quadratic Discriminant Analysis Based Spiral Dynamic Optimization Algorithm

Classification of cognitive impairment using electroencephalography for clinical inspection.

Impairment in cognitive skill though set-in due to various diseases, its progress is based on neuronal degeneration. In general, cognitive impairment (CI) is divided into three stages: mild, moderate and severe. Quantification of CI is important for deciding/changing therapy. Attempted in this work is to quantify electroencephalograph (EEG) signal and group it into four classes (controls and three stages of CI). After acquiring resting state EEG signal from the participants, non-local and local synchrony measures are derived from phase amplitude coupling and phase locking value. This totals to 160 features per individual for each task. Two types of classification networks are constructed. The first one is an artificial neural network (ANN) that takes derived features and gives a maximum accuracy of 85.11%. The second network is convolutional neural network (CNN) for which topographical images constructed from EEG features becomes the input dataset. The network is trained with 60% of data and then tested with remaining 40% of data. This process is performed in 5-fold technique, which yields an average accuracy of 94.75% with only 30 numbers of inputs for every individual. The result of the study shows that CNN outperforms ANN with a relatively lesser number of inputs. From this it can be concluded that this method proposes a simple task for acquiring EEG (which can be done by CI subjects) and quantifies CI stages with no overlapping between control and test group, thus making it possible for identifying early symptoms of CI.

Clinical classification of memory and cognitive impairment with multimodal digital biomarkers.

Early detection of Alzheimer's disease and cognitive impairment is critical to improving the healthcare trajectories of aging adults, enabling early intervention and potential prevention of decline. To evaluate multi-modal feature sets for assessing memory and cognitive impairment, feature selection and subsequent logistic regressions were used to identify the most salient features in classifying Rey Auditory Verbal Learning Test-determined memory impairment. Multimodal models incorporating graphomotor, memory, and speech and voice features provided the stronger classification performance (area under the curve=0.83; sensitivity=0.81, specificity=0.80). Multimodal models were superior to all other single modality and demographics models. The current research contributes to the prevailing multimodal profile of those with cognitive impairment, suggesting that it is associated with slower speech with a particular effect on the duration, frequency, and percentage of pauses compared to normal healthy speech.

MRI-based mild cognitive impairment and Alzheimer's disease classification using an algorithm of combination of variational autoencoder and other machine learning classifiers

Background Correctly diagnosing mild cognitive impairment (MCI) and Alzheimer's disease (AD) is important for patient selection in drug discovery. Research outcomes on stage diagnosis using neuroimages combined with cerebrospinal fluid and genetic biomarkers are expensive and time-consuming. Only structural magnetic resonance imaging (sMRI) scans from two internationally recognized datasets are employed as input as well as test and independent validation to determine the classification of dementia by the machine learning algorithm. Objective We extract the reduced dimensional latent feature vector from the sMRI scans using a variational autoencoder (VAE). The objective is to classify AD, MCI, and control (CN) using MRI and without any other information. Methods The extracted feature vectors from MRI scans by VAE are used as input conditions for different advanced machine-learning classifiers. Classification of AD/CN/MCI are conducted using the output of VAE from MRI images and different artificial intelligence/machine learning classifier models in two cohorts. Results Using only MRI scans, the primary goal of the study is to test the ability to classify AD from CN and MCI cases. The current study achieved classification accuracies of AD versus CN 75.45% (F1-score = 79.52%), AD versus MCI 81.41% (F1-Score = 87.06%), and autopsy-confirmed AD versus MCI 92.75% (F1-Score = 95.52%) in test sets and AD versus CN 86.16% (F1-score = 92.03%) and AD versus MCI 70.03% (F1-Score = 82.1%) in validation data set. Conclusions By overcoming the data leakage problem, the autopsy-confirmed machine learning classification model is tested in two independent cohorts. External validation by an independent cohort improved the quality and novelty of the classification algorithm.

Classification of cognitive impairment in older adults based on brain functional state measurement data via hierarchical clustering analysis.

Cognitive impairment (CI) is a common degenerative condition in the older population. However, the current methods for assessing CI are not based on brain functional state, which leads to delayed diagnosis, limiting the initiatives towards achieving early interventions. A total of one hundred and forty-nine community-dwelling older adults were recruited. Montreal Cognitive Assessment (MoCA) and Mini-Mental State Exam (MMSE) were used to screen for CI, while brain functional was assessed by brain functional state measurement (BFSM) based on electroencephalogram. Bain functional state indicators associated with CI were selected by lasso and logistic regression models (LRM). We then classified the CI participants based on the selected variables using hierarchical clustering analysis. Eighty-one participants with CI detected by MoCA were divided into five groups. Cluster 1 had relatively lower brain functional states. Cluster 2 had highest mental task-switching index (MTSi, 13.7 ± 3.4), Cluster 3 had the highest sensory threshold index (STi, 29.9 ± 7.7), Cluster 4 had high mental fatigue index (MFi) and cluster 5 had the highest mental refractory period index (MRPi), and external apprehension index (EAi) (21.6 ± 4.4, 35.4 ± 17.7, respectively). Thirty-three participants with CI detected by MMSE were divided into 3 categories. Cluster 1 had the highest introspective intensity index (IIi, 63.4 ± 20.0), anxiety tendency index (ATi, 67.2 ± 13.6), emotional resistance index (ERi, 50.2 ± 11.9), and hypoxia index (Hi, 41.8 ± 8.3). Cluster 2 had the highest implicit cognitive threshold index (ICTi, 87.2 ± 12.7), and cognitive efficiency index (CEi, 213.8 ± 72.0). Cluster 3 had higher STi. The classifications both showed well intra-group consistency and inter-group variability. In our study, BFSM-based classification can be used to identify clinically and brain-functionally relevant CI subtypes, by which clinicians can perform personalized early rehabilitation.

Sensitivity of Individual and Composite Test Scores from the Cogstate Brief Battery to Mild Cognitive Impairment and Dementia Due to Alzheimer's Disease.

The Cogstate Brief Battery (CBB) is a computerized cognitive test battery used commonly to identify cognitive deficits related to Alzheimer's disease (AD). However, AD and normative samples used to understand the sensitivity of the CBB to AD in the clinic have been limited, as have the outcome measures studied. This study investigated the sensitivity of CBB outcomes, including potential composite scores, to cognitive impairment in mild cognitive impairment (MCI) and dementia due to AD, in carefully selected samples. Samples consisted of 4,871 cognitively unimpaired adults and 184 adults who met clinical criteria for MCI (Clinical Dementia Rating (CDR) = 0.5) or dementia (CDR > 0.5) due to AD and CBB naive. Speed and accuracy measures from each test were examined, and theoretically- and statistically-derived composites were created. Sensitivity and specificity of classification of cognitive impairment were compared between outcomes. Individual CBB measures of learning and working memory showed high discriminability for AD-related cognitive impairment for CDR 0.5 (AUCs ∼ 0.79-0.88), and CDR > 0.5 (AUCs ∼ 0.89-0.96) groups. Discrimination ability for theoretically derived CBB composite measures was high, particularly for the Learning and Working Memory (LWM) composite (CDR 0.5 AUC = 0.90, CDR > 0.5 AUC = 0.97). As expected, statistically optimized linear composite measures showed strong discrimination abilities albeit similar to the LWM composite. In older adults, the CBB is effective for discriminating cognitive impairment due to MCI or AD-dementia from unimpaired cognition with the LWM composite providing the strongest sensitivity.

EEG-Based Functional Connectivity Analysis for Cognitive Impairment Classification

Understanding how mild cognitive impairment affects global neural networks may explain changes in brain electrophysiology. Using graph theory and the visual oddball paradigm, we evaluated the functional connectivity of neuronal networks in brain lobes. The study involved 30 participants: 14 with mild cognitive impairment (MCI) and 16 healthy control (HC) participants. We conducted an examination using the visual oddball paradigm, focusing on electroencephalography signals with targeted stimuli. Our analysis employed functional connectivity utilizing the change point detection method. Additionally, we implemented training for linear discriminant analysis, K-nearest neighbor, and decision tree techniques to classify brain activity, distinguishing between subjects with mild cognitive impairment and those in the healthy control group. Our results demonstrate the efficacy of combining functional connectivity measurements derived from electroencephalography with machine learning for cognitive impairment classification. This research opens avenues for further exploration, including the potential for real-time detection of cognitive decline in complex real-world scenarios.

Early rise of glutamate-glutamine levels in mild cognitive impairment: Evidence for emerging excitotoxicity

BackgroundUse proton magnetic resonance spectroscopy (1H-MRS) non invasive technique to assess the modifications of glutamate-glutamine (Glx) and gammaaminobutyric acid (GABA) brain levels in patients reporting a cognitive complain MethodsPosterior cingular cortex 1H-MRS spectra of 46 patients (19 male, 27 female) aged 57 to 87 years (mean : 73.32 ± 7.33 years) with a cognitive complaint were examined with a MEGA PRESS sequence at 3T, and compounds Glutamateglutamine (Glx), GABA, Creatine (Cr) and NAA were measured. From this data the metabolite ratios Glx/Cr, GABA/Cr and NAA/Cr were calculated. In addition, all patient performed the Mini Mental State Evaluation (MMSE) and 2 groups were realized with the clinical threshold of 24. Results16 patients with MMSE 〈 24 and 30 patients with MMSE 〉 24. Significant increase of Glx/Cr in PCC of patients with MMSE 〈 24 compared to patients with MMSE 〉 24. Moreover, GABA/Cr ratio exhibited a trend for a decrease in PCC between the two groups, while they showed a significant decrease NAA/Cr ratio. ConclusionOur results concerning Glx are in agreement with a physiopathological hypothesis involving a biphasic variation of glutamate levels associated with excitotoxicity, correlated with the clinical evolution of the disease. These observations suggest that MRS assessment of glutamate levels could be helpful for both diagnosis and classification of cognitive impairment in stage.

Walking path images from real-time location data predict degree of cognitive impairment

BackgroundWe propose a novel approach that uses spatial walking patterns produced by real-time location systems to classify the severity of cognitive impairment (CI) among residents of a memory care unit. MethodsEach participant was classified as “No-CI”, “Mild-Moderate CI” or “Severe CI” based on their Mini-Mental State Examination scores. The location data was distributed into windows of various durations (5, 10, 15 and 30 min) and transformed into images used to train a custom convolutional neural network (CNN) at each window size. Class Activation Mapping was applied to the top-performing models to determine the features of images associated with each class. ResultsThe best performing model achieved an accuracy of 87.38 % (30-min window length) with an overall pattern that larger window sizes perform better. The class activation maps were effectively consolidated into a Cognitive Impairment Classification Value (CICV) score that distinguishes between No-CI, Mild-Moderate CI, and Severe CI. ConclusionThe class activation maps show that the CNN made relevant and intuitive distinctions for paths corresponding to each class. Future work should validate the proposed techniques with participants who are well-characterized clinically, over larger and diversified settings, and towards classification of neuropsychiatric symptoms such as motor agitation, mood, or apathy.

Cognitive impairment in people living with HIV: consensus recommendations for a new approach.

Current approaches to classifying cognitive impairment in people living with HIV can overestimate disease burden and lead to ambiguity around disease mechanisms. The 2007 criteria for HIV-associated neurocognitive disorders (HAND), sometimes called the Frascati criteria, can falsely classify over 20% of cognitively healthy individuals as having cognitive impairment. Minimum criteria for HAND are met on the basis of performance on cognitive tests alone, which might not be appropriate for populations with diverse educational and socioeconomic backgrounds. Imprecise phenotyping of cognitive impairment can limit mechanistic research, biomarker discovery and treatment trials. Importantly, overestimation of cognitive impairment carries the risk of creating fear among people living with HIV and worsening stigma and discrimination towards these individuals. To address this issue, we established the International HIV-Cognition Working Group, which is globally representative and involves the community of people living with HIV. We reached consensus on six recommendations towards a new approach for diagnosis and classification of cognitive impairment in people living with HIV, intended to focus discussion and debate going forward. We propose the conceptual separation of HIV-associated brain injury - including active or pretreatment legacy damage - from other causes of brain injury occurring in people living with HIV. We suggest moving away from a quantitative neuropsychological approach towards an emphasis on clinical context. Our recommendations are intended to better represent the changing profile of cognitive impairment in people living with HIV in diverse global settings and toprovide a clearer framework of classification for clinical management and research studies.

Behavioral and biologic characteristics of cancer-related cognitive impairment biotypes

Psychiatric diagnosis is moving away from symptom-based classification and towards multi-dimensional, biologically-based characterization, or biotyping. We previously identified three biotypes of chemotherapy-related cognitive impairment based on functional brain connectivity. In this follow-up study of 80 chemotherapy-treated breast cancer survivors and 80 non-cancer controls, we evaluated additional factors to help explain biotype expression: neurofunctional stability, brain age, apolipoprotein (APOE) genotype, and psychoneurologic symptoms. We also compared the discriminative ability of a traditional, symptom-based cognitive impairment definition with that of biotypes. We found significant differences in cortical brain age (F = 10.50, p < 0.001), neurofunctional stability (F = 2.83, p = 0.041), APOE e4 genotype (X2 = 7.68, p = 0.050), and psychoneurological symptoms (Pillai = 0.378, p < 0.001) across the three biotypes. The more resilient Biotype 2 demonstrated significantly higher neurofunctional stability compared to the other biotypes. Symptom-based classification of cognitive impairment did not differentiate biologic or other behavioral variables, suggesting that traditional categorization of cancer-related cognitive effects may miss important characteristics which could inform targeted treatment strategies. Additionally, biotyping, but not symptom-typing, was able to distinguish survivors with cognitive versus psychological effects. Our results suggest that Biotype 1 survivors might benefit from first addressing symptoms of anxiety and fatigue, Biotype 3 might benefit from a treatment plan which includes sleep hygiene, and Biotype 2 might benefit most from cognitive skills training or rehabilitation. Future research should include additional demographic and clinical information to further investigate biotype expression related to risk and resilience and examine integration of more clinically feasible imaging approaches.

Sensory impairment and algorithmic classification of early cognitive impairment.

Sensory impairment (SI) is linked to cognitive decline, but its association with early cognitive impairment (ECI) is unclear. Sensory functions (vision, hearing, vestibular function, proprioception, and olfaction) were measured between 2012 and 2018 in 414 Baltimore Longitudinal Study of Aging (BLSA) participants (age 74±9 years; 55% women). ECI was defined as 1 standard deviation below age-, sex-, race-, and education-specific mean performance in Card Rotations or California Verbal Learning Test immediate recall. Log binomial models (cross-sectional analysis) and Cox regression models (time-to-event analysis) were used to examine the association between SI and ECI. Cross-sectionally, participants with ≥3 SI had twice the prevalence of ECI (prevalence ratio=2.10, p=0.02). Longitudinally, there was no significant association between SI and incident ECI over up to 6 years of follow-up. SI is associated with higher prevalence, but not incident ECI. Future studies with large sample sizes need to further elucidate the relationship between SI and ECI. Sensory impairment is associated with high prevalence of early cognitive impairmentMultisensory impairment may pose a strong risk of early changes in cognitive functionIdentifying multisensory impairment may help early detection of dementia.

Measurement differences in the assessment of functional limitations for cognitive impairment classification across geographic locations.

The measurement of dementia in cross-national contexts relies on the assessment of functional limitations. We aimed to evaluate the performance of survey items on functional limitations across culturally diverse geographic settings. We used data from the Harmonized Cognitive Assessment Protocol Surveys (HCAP) in five countries (total N=11,250) to quantify associations between items on functional limitations and cognitive impairment. Many items performed better in the United States and England compared to South Africa, India, and Mexico. Items on the Community Screening Instrument for Dementia (CSID) had the least variability across countries (SD=0.73vs. 0.92 [Blessed] and 0.98 [Jorm IQCODE]), but also the weakest associations with cognitive impairment (median odds ratio [OR]=2.23vs. 3.01 [Blessed] and 2.75 [Jorm IQCODE]). Differences in cultural norms for reporting functional limitations likely influences performance of items on functional limitations and may affect the interpretation of results from substantive studies. There was substantial cross-country variation in item performance. Items from the Community Screening Instrument for Dementia (CSID) had less cross-country variability but lower performance. There was more variability in performance of instrumental activities of daily living (IADL) compared to activities of daily living (ADL) items. Variability in cultural expectations of older adults should be taken into account. Results highlight the need for novel approaches to assessing functional limitations.

A Convolutional Neural Network and Graph Convolutional Network Based Framework for AD Classification.

The neuroscience community has developed many convolutional neural networks (CNNs) for the early detection of Alzheimer's disease (AD). Population graphs are thought of as non-linear structures that capture the relationships between individual subjects represented as nodes, which allows for the simultaneous integration of imaging and non-imaging information as well as individual subjects' features. Graph convolutional networks (GCNs) generalize convolution operations to accommodate non-Euclidean data and aid in the mining of topological information from the population graph for a disease classification task. However, few studies have examined how GCNs' input properties affect AD-staging performance. Therefore, we conducted three experiments in this work. Experiment 1 examined how the inclusion of demographic information in the edge-assigning function affects the classification of AD versus cognitive normal (CN). Experiment 2 was designed to examine the effects of adding various neuropsychological tests to the edge-assigning function on the mild cognitive impairment (MCI) classification. Experiment 3 studied the impact of the edge assignment function. The best result was obtained in Experiment 2 on multi-class classification (AD, MCI, and CN). We applied a novel framework for the diagnosis of AD that integrated CNNs and GCNs into a unified network, taking advantage of the excellent feature extraction capabilities of CNNs and population-graph processing capabilities of GCNs. To learn high-level anatomical features, DenseNet was used; a set of population graphs was represented with nodes defined by imaging features and edge weights determined by different combinations of imaging or/and non-imaging information, and the generated graphs were then fed to the GCNs for classification. Both binary classification and multi-class classification showed improved performance, with an accuracy of 91.6% for AD versus CN, 91.2% for AD versus MCI, 96.8% for MCI versus CN, and 89.4% for multi-class classification. The population graph's imaging features and edge-assigning functions can both significantly affect classification accuracy.