Search

Qiu et al. present a novel deep learning strategy to generate high-resolution visualizations of Alzheimer’s disease risk in humans that are highly interpretable and can accurately predict Alzheimer’s disease status. They then test the model by comparing its performance to that of neurologists and neuropathological data., Alzheimer’s disease is the primary cause of dementia worldwide, with an increasing morbidity burden that may outstrip diagnosis and management capacity as the population ages. Current methods integrate patient history, neuropsychological testing and MRI to identify likely cases, yet effective practices remain variably applied and lacking in sensitivity and specificity. Here we report an interpretable deep learning strategy that delineates unique Alzheimer’s disease signatures from multimodal inputs of MRI, age, gender, and Mini-Mental State Examination score. Our framework linked a fully convolutional network, which constructs high resolution maps of disease probability from local brain structure to a multilayer perceptron and generates precise, intuitive visualization of individual Alzheimer’s disease risk en route to accurate diagnosis. The model was trained using clinically diagnosed Alzheimer’s disease and cognitively normal subjects from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset (n = 417) and validated on three independent cohorts: the Australian Imaging, Biomarker and Lifestyle Flagship Study of Ageing (AIBL) (n = 382), the Framingham Heart Study (n = 102), and the National Alzheimer’s Coordinating Center (NACC) (n = 582). Performance of the model that used the multimodal inputs was consistent across datasets, with mean area under curve values of 0.996, 0.974, 0.876 and 0.954 for the ADNI study, AIBL, Framingham Heart Study and NACC datasets, respectively. Moreover, our approach exceeded the diagnostic performance of a multi-institutional team of practicing neurologists (n = 11), and high-risk cerebral regions predicted by the model closely tracked post-mortem histopathological findings. This framework provides a clinically adaptable strategy for using routinely available imaging techniques such as MRI to generate nuanced neuroimaging signatures for Alzheimer’s disease diagnosis, as well as a generalizable approach for linking deep learning to pathophysiological processes in human disease.

Quantifying heterogeneity in Alzheimer’s disease (AD) risk is critical for individualized care and management. Recent attempts to assess AD heterogeneity have used structural (magnetic resonance imaging (MRI)-based) or functional (Aβ or tau) imaging, which focused on generating quartets of atrophy patterns and protein spreading, respectively. Here we present a computational framework that facilitated the identification of subtypes based on their risk of progression to AD. We used cerebrospinal fluid (CSF) measures of Aβ from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) (n=544, discovery cohort) as well as the National Alzheimer’s Coordinating Center (NACC) (n=508, validation cohort), and risk-stratified individuals with mild cognitive impairment (MCI) into quartiles (high-risk (H), intermediate-high risk (IH), intermediate-low risk (IL), and low-risk (L)). Patients were divided into subgroups utilizing patterns of brain atrophy found in each of these risk-stratified quartiles. We found H subjects to have a greater risk of AD progression compared to the other subtypes at 2- and 4-years in both the discovery and validation cohorts (ADNI: H subtype versus all others, p < 0.05 at 2 and 4 years; NACC: H vs. IL and LR at 2 years, p < 0.05, and a trend toward higher risk vs. IH, and p < 0.05 vs. IH, and L risk groups at 48 months with a trend toward lower survival vs. IL). Using MRI-based neural models that fused various deep neural networks with survival analysis, we then predicted MCI to AD conversion. We used these models to identify subtype-specific regions that demonstrate the largest levels of atrophy-related importance, which had minimal overlap (Average pairwise Jaccard Similarity in regions between the top 5 subtypes, 0.25±0.05 (± std)). Neuropathologic changes characteristic of AD were present across all subtypes in comparable proportions (Chi-square test, p>0.05 for differences in ADNC, n=31). Our risk-based approach to subtyping individuals provides an objective means to intervene and tailor care management strategies at early stages of cognitive decline.

Introduction Despite the availability of age- and education-adjusted standardized scores for most neuropsychological tests, there is a lack of objective rules in how to interpret multiple concurrent neuropsychological test scores that characterize the heterogeneity of Alzheimer's disease. Methods Using neuropsychological test scores of 2091 participants from the Framingham Heart Study, we devised an automated algorithm that follows general diagnostic criteria and explores the heterogeneity of Alzheimer's disease. Results We developed a series of stepwise diagnosis rules that evaluate information from multiple neuropsychological tests to produce an intuitive and objective Alzheimer's disease dementia diagnosis with more than 80% accuracy. Discussion A data-driven stepwise diagnosis system is useful for diagnosis of Alzheimer's disease from neuropsychological tests. It demonstrated better performance than the traditional dichotomization of individuals' performance into satisfactory and unsatisfactory outcomes, making it more reflective of dementia as a spectrum disorder. This algorithm can be applied to both within clinic and outside-of-clinic settings., Highlights • Stepwise diagnosis rules for Alzheimer's disease (AD) dementia diagnosis. • Heterogeneous AD profiles based on specific AD risk factors identified. • Selection of important neuropsychological tests aid differential diagnoses. • Grading system more reflective of dementia as a spectrum disorder. • Automated diagnosis rules applicable to both within clinic and outside-of-clinic settings.

Additional file 1: Figure S1. (A) Simple GAN architecture with a generator and a discriminator. (B) Simple FCN classification architecture for prediction of AD status. Figure S2. Number of cases from the ADNI cohort used for GAN and FCN model development. The cases covered by the black arrow indicate the ones used for the GAN model development and the cases covered by the red arrow indicate the ones used for the FCN model development. Figure S3. ANCOVA analysis to assess the mean image quality using SNR, BRISQUE and NIQE in the ADNI-training (a-c) and ADNI-validation (d-f) data. Table S1. Specific group differences in magnetic field strength (1.5T, 3T, 3T*) evaluated using Tukey's post hoc procedure across the image quality measures SNR, BRISQUE and NIQE in the ADNI-training, testing and validation data. Table S2. ANOVA analysis of AUCs from the ROC curves on the gender, age and scanner variables to demonstrate whether the AD classification performance differed on various groups. Figure S4. Performance of the FCN classifier based on the images generated using the simpleGAN architecture (Figure S1). (A) Sensitivity-specificity and (B) precision-recall curves are shown on the ADNI test, AIBL and NACC datasets, respectively. Figure S5. ANCOVA analysis to assess the mean image quality using SNR, BRISQUE and NIQE for simpleGAN model in the ADNI-test (a-c) and t-test results of NACC (d-f) and AIBL (g-i) data, respectively. Figure S6. ANCOVA analysis to assess the mean image quality using SNR, BRISQUE and NIQE for simpleGAN model in the ADNI-training (a-c) and ADNI-validation (d-f) data. The effect of independent variables such as age, education, MMSE scores, and type of scanner was evaluated using a stepwise forward selection process and models were adjusted accordingly. Table S3. Specific group differences in magnetic field strength (1.5T, 3T, 3T*) were evaluated using Tukey's post hoc procedure across the image quality measures SNR, BRISQUE and NIQE for the simpleGAN model in the ADNI-training, testing and validation data.

Background Generative adversarial networks (GAN) can produce images of improved quality but their ability to augment image-based classification is not fully explored. We evaluated if a modified GAN can learn from magnetic resonance imaging (MRI) scans of multiple magnetic field strengths to enhance Alzheimer’s disease (AD) classification performance. Methods T1-weighted brain MRI scans from 151 participants of the Alzheimer’s Disease Neuroimaging Initiative (ADNI), who underwent both 1.5-Tesla (1.5-T) and 3-Tesla imaging at the same time were selected to construct a GAN model. This model was trained along with a three-dimensional fully convolutional network (FCN) using the generated images (3T*) as inputs to predict AD status. Quality of the generated images was evaluated using signal to noise ratio (SNR), Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) and Natural Image Quality Evaluator (NIQE). Cases from the Australian Imaging, Biomarker & Lifestyle Flagship Study of Ageing (AIBL, n = 107) and the National Alzheimer’s Coordinating Center (NACC, n = 565) were used for model validation. Results The 3T*-based FCN classifier performed better than the FCN model trained using the 1.5-T scans. Specifically, the mean area under curve increased from 0.907 to 0.932, from 0.934 to 0.940, and from 0.870 to 0.907 on the ADNI test, AIBL, and NACC datasets, respectively. Additionally, we found that the mean quality of the generated (3T*) images was consistently higher than the 1.5-T images, as measured using SNR, BRISQUE, and NIQE on the validation datasets. Conclusion This study demonstrates a proof of principle that GAN frameworks can be constructed to augment AD classification performance and improve image quality.

BackgroundGenerative adversarial networks (GAN) can generate images of improved quality but their ability to augment image-based classification tasks is not fully explored.PurposeWe evaluated if a modified GAN can learn from MRI scans of multiple magnetic field strength to enhance Alzheimer’s disease (AD) classification performance.Materials and methodsT1-weighted brain MRI scans from 151 participants of the Alzheimer’s Disease Neuroimaging Initiative (ADNI), who underwent both 1.5 Tesla (1.5T) and 3 Tesla imaging at the same time were selected to construct a GAN model. This model was trained along with a three-dimensional fully convolutional network (FCN) using the generated images (1.5T*) as inputs to predict AD status. Quality of the generated images was evaluated using signal to noise ratio (SNR), Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) and Natural Image Quality Evaluator (NIQE). Data from the Australian Imaging, Biomarker & Lifestyle Flagship Study of Ageing (AIBL, n=107), and the National Alzheimer’s Coordinating Center (NACC, n=565) was used for model validation.ResultsThe mean quality of the generated (1.5T*) images was consistently higher than the 1.5T images, as measured using SNR, BRISQUE and NIQE on the validation datasets. The 1.5T*-based FCN classifier performed better than the FCN model trained using the 1.5T scans.Specifically, the mean area under curve increased from 0.907 to 0.932, from 0.934 to 0.940 and from 0.870 to 0.907 on the ADNI test, AIBL and NACC datasets, respectively.ConclusionThis study demonstrates that GAN frameworks can be constructed to simultaneously improve image quality and augment classification performance.Key pointsOur proposed generative adversarial network used 1.5 and 3 Tesla MR scans of the brain to generate images of improved quality, as estimated using no-reference image quality algorithms.Classification models of Alzheimer’s disease risk developed using the generated images had higher classification performance than the models developed using the original 1.5 Tesla scans.

Introduction Subtle cognitive alterations that precede clinical evidence of cognitive impairment may help predict the progression to Alzheimer’s disease (AD). Neuropsychological (NP) testing is an attractive modality for screening early evidence of AD. Methods Longitudinal NP and demographic data from the Framingham Heart Study (FHS; N = 1696) and the National Alzheimer's Coordinating Center (NACC; N = 689) were analyzed using an unsupervised machine learning framework. Features, including age, logical memory-immediate and delayed recall, visual reproduction-immediate and delayed recall, the Boston naming tests, and Trails B, were identified using feature selection, and processed further to predict the risk of development of AD. Results Our model yielded 83.07 ± 3.52% accuracy in FHS and 87.57 ± 1.19% accuracy in NACC, 80.52 ± 3.93%, 86.74 ± 1.63% sensitivity in FHS and NACC respectively, and 85.63 ± 4.71%, 88.41 ± 1.38% specificity in FHS and NACC, respectively. Discussion Our results suggest that a subset of NP tests, when analyzed using unsupervised machine learning, may help distinguish between high- and low-risk individuals in the context of subsequent development of AD within 5 years. This approach could be a viable option for early AD screening in clinical practice and clinical trials.

The Framingham Heart Study (FHS) was established in 1948 and is the longest, ongoing prospective cohort study studying cardiovascular diseases and stroke. Initially, the FHS was primarily focused on cardiovascular diseases; however, over recent decades data from the FHS have found that lifetime exposure to major cardiovascular risk factors such as hypertension, diabetes mellitus, arterial stiffness, and pressure pulsatility adversely impact neuropsychological functioning particularly in older individuals contributing to the onset of dementia including Alzheimer’s disease. Research from the FHS suggest that appropriate management of cardiovascular risk factors as well as healthy lifestyle practices that include regular blood pressure monitoring, not smoking, and eating a choline-rich diet appear to minimize neuropsychological decline as well as dementia onset.

The relationship between body mass index (BMI) and health outcomes of older adults, including dementia, remains controversial. Many studies find inverse associations between BMI and dementia among older adults, while in other studies high BMI in midlife is associated with increased dementia risk. In this issue, Li et al. (Am J Epidemiol. 2021;190(12):2503–2510) examine BMI from mid- to late life and risk of dementia using the extensive follow-up of the Framingham Offspring Study. They found changing trends in the association between BMI and dementia from a positive association for midlife (ages 40–49) to an inverse trend in late life. Their work demonstrates the importance of studying dementia risk factors across the life course. Midlife obesity might be an important modifiable risk factor for dementia. However, because incipient dementia can lead to weight loss, reverse causation remains a key source of bias that could explain an inverse trend between BMI and dementia in older ages. The extent of other biases, including unmeasured confounding, inaccuracy of BMI as a measure for adiposity, or selective survival, are also unclear. Triangulating evidence on body composition and dementia risk could lead to better targets for dementia intervention, but future work will need to evaluate specific pathways.

Alzheimer’s disease (AD) is the primary cause of dementia worldwide (1), with an increasing morbidity burden that may outstrip diagnosis and management capacity as the population ages. Current methods integrate patient history, neuropsychological testing and magnetic resonance imaging (MRI) to identify likely cases, yet effective practices remain variably-applied and lacking in sensitivity and specificity (2). Here we report an explainable deep learning strategy that delineates unique AD signatures from multimodal inputs of MRI, age, gender, and mini-mental state examination (MMSE) score. Our framework linked a fully convolutional network (FCN) to a multilayer perceptron (MLP) to construct high resolution maps of disease probability from local brain structure. This enabled precise, intuitive visualization of individual AD risk en route to accurate diagnosis. The model was trained using clinically-diagnosed AD and cognitively normal (NC) subjects from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset (n=417) (3), and validated on three independent cohorts: the Australian Imaging, Biomarker & Lifestyle Flagship Study of Ageing (AIBL, n=382) (4), the Framingham Heart Study (FHS, n=102) (5), and the National Alzheimer’s Coordinating Center (NACC, n=582) (6). Model performance was consistent across datasets, with mean accuracy values of 0.966, 0.948, 0.815, and 0.916 for ADNI, AIBL, FHS and NACC, respectively. Moreover, our approach exceeded the diagnostic performance of a multi-institutional team of practicing neurologists (n=11), and high-risk cerebral regions predicted by the model closely tracked postmortem histopathological findings. This framework provides a clinically-adaptable strategy for using routinely available imaging techniques such as MRI to generate nuanced neuroimaging signatures for AD diagnosis, as well as a generalizable approach for linking deep learning to pathophysiological processes in human disease.

Introduction Because many older adults lack dental insurance and have limited or no access to dental care, it is essential to train future physicians to conduct brief oral health assessments on them. Likewise, interprofessional educational experiences are crucial in teaching medical students the skills necessary to provide comprehensive, team-based care to complex and vulnerable populations. Thus, this workshop was designed to increase fourth-year medical students' knowledge and confidence in performing oral health examinations on older adult patients using an interprofessional and hands-on approach. Methods The curriculum includes an online presession self-study module followed by a 75-minute workshop. The workshop is comprised of a brief introduction (5 minutes), a lecture about the impact of oral health on older adults (30 minutes), a hands-on skill session practicing a focused oral exam led by dental students (30 minutes), and a large-group debrief and wrap-up (10 minutes). Results A pre-/postsurvey assessed learners' knowledge, attitudes, and confidence in oral health skills. The results were compared to a lecture-only format that was in place prior to the implementation of the workshop. In comparison to medical students who received the lecture-only format, those who participated in the workshop showed a greater increase in confidence and skills over time. The medical students also expressed interest in more frequent opportunities for collaborative learning experiences with dental students. Discussion This workshop was successful in introducing an interprofessional experience to medical students in order to ensure more comprehensive and coordinated care for older adult patients in the future.