Digital speech-based assessments provide scalable tools for detecting subtle cognitive decline. Here, we investigated whether digitally derived speech-based composite score of cognition and individual speech features were associated with alterations in functional connectivity (FC) within task-related brain networks in the Alzheimer’s disease spectrum, which are known to reflect cognitive performance and disease-related changes. Data were analyzed from 129 participants of the German PROSPECT-AD study, ranging from cognitively healthy individuals to those with mild cognitive impairment. Speech-based cognitive scores and speech features were derived from automated phone-administered semantic verbal fluency (SVF) and verbal learning tasks (VLT). Resting-state fMRI assessed FC, with intrinsic connectivity networks identified via independent component analysis and dual regression. Associations were examined using permutation-based voxel-wise regression, controlling for demographic and clinical covariates. Seed-to-voxel analyses were conducted to support network identification and complement findings. Greater language network connectivity in the left middle temporal gyrus was associated with increased SVF temporal cluster switching (FWE < .05, cluster size = 12 voxels, mean T = 3.86). Exploratory analyses (uncorrected p < .01) demonstrated no significant associations between cognitive composite scores and FC. However, individual SVF and VLT speech features exhibited network-specific associations across executive, language, and default mode networks, indicating exploratory yet spatially distinct connectivity patterns. Digital speech-based assessments may have limited current utility for detecting FC alterations in at-risk individuals. Further validation using complementary methodological approaches, shorter intervals between fMRI and speech assessments, and testing in independent cohorts, are essential to establish their reliability and clinical relevance for monitoring brain network changes.

Background: Artificial intelligence (AI) is expected to become increasingly important in hospitals and clinical care. However, successful implementation depends on technical, organizational, and user-related factors. We aimed to identify the key requirements for equitable and effective adoption of AI-based applications in clinical practice. Methods: Within the “Clinical AI-Based Diagnostics” (CAIDX) project, we conducted systematic expert interviews with 51 stakeholders from four European countries working across the healthcare sector. Interviews explored the perceived added value of AI, barriers to implementation, usability requirements, and potential health-economic implications. Findings: Stakeholder analysis revealed several recurrent sources of tension affecting implementation: limited physician involvement, differing perspectives between IT specialists and clinicians, unclear distribution of tasks and responsibilities, and low institutional commitment driven by concerns about inefficient use of time and resources. We also identified structural and workforce-related constraints that make implementation particularly challenging, including the large number of actors involved and the highly individual organizational setup of hospitals, which limits transferability of implementation models. Interpretation: Implementation of clinical AI requires targeted action at hospital, vendor, and health-system levels. Compared with earlier healthcare technologies, AI adoption is shaped more strongly by AI literacy, professional attitudes, and perceptions of trust and responsibility. Based on these findings, we propose practical recommendations to support implementation of AI applications in routine care.

IntroductionNormal aging is associated with alterations of functional connectivity in brain neuronal networks. Altered network connectivity may be associated with accelerated cognitive decline. Physical activity is considered a beneficial lifestyle factor for maintaining cognitive health. Higher intensities of physical activity may induce structural and functional changes in the brain, particularly in regions involved in cognitive functions. However, the underlying neural mechanisms are not widely investigated. Our aim was to examine the association between resting-state functional connectivity of brain networks previously associated with cognitive and motor functions, physical activity and cognitive performance in healthy older adults.MethodsWe analyzed resting-state fMRI, physical activity and neuropsychological data of 149 healthy older adults (mean age: 68 years). Physical activity was measured by using actigraphs worn for 7 days and categorized into moderate-to-vigorous activity. Euclidean norm minus one values used to represent mean overall physical activity. We used a hypothesis driven seed-based approach and data-driven independent component analysis to examine brain network activity of a priori selected brain regions and networks.ResultsNo significant associations were found in the seed-based analyses. The independent component analyses showed spatially restricted effects of moderate-to-vigorous physical activity in frontal regions of the default mode and salience networks, at p < 0.01 uncorrected.ConclusionDifferent physical activity intensities were not significantly associated with resting-state functional connectivity of various brain networks in a sample of healthy older adults. This finding contrasts with the results of previous cross-sectional studies.

We propose five recommendations to make AI-based research studies more suitable for a clinical readership. First, authors should justify the added value of complex and potentially more opaque AI approaches. Second, rigorous description of input data, diagnostic criteria, and preprocessing is essential to avoid biased or clinically irrelevant outcomes. Third, benchmarking against clinically relevant performance thresholds should be established a priori. Fourth, method sections should combine an accessible lay summary with detailed technical supplement. Fifth, model explainability is encouraged to mitigate opacity. These recommendations aim to support AI research that is methodologically robust and interpretable for AD researchers.

It is well-known that in patients with Alzheimer's disease (AD) and high education attainment, cognitive performance is typically better than expected based on the burden of brain neuropathology and neurodegeneration (Stern et al., 2018; doi: 10.1016/j.neuroimage.2018.05.033). This resilience of the cognitive status was attributed to a sort of cognitive reserve (CR) accumulated by persons with high education attainment, which predicts a life with engaging job, intellectual, and social demands (Arenaza-Urquijo et al. 2015; doi: 10.3389/fnagi.2015.00134; Stern et al. 2018). Previous resting-state eyes-closed electroencephalographic (rsEEG) studies showed that alpha rhythms in posterior visual and visuospatial areas are related to CR in healthy adults, subjective memory complaints (SMC) seniors, and patients with mild cognitive impairment due to AD (ADMCI). In the present exploratory study, we used the database of the INSIGHT cohort (Dubois et al., 2018; doi: 10.1016/S1474-4422(18)30029-2), we investigated whether older adults with subjective memory complaints (SMC) and brain amyloid-β accumulation may exhibit clinical progression over 2 years as a function of educational attainment (a proxy of cognitive reserve). SMCneg with high educational attainment (Edu+) participants showed greater posterior rsEEG alpha rhythms compared to SMCneg with low educational attainment (Edu-) participants. In contrast, SMCpos Edu+ participants exhibited reduced posterior rsEEG alpha rhythms and parietal cortical thickness compared to SMCpos Edu- participants. No EEG (Figure 1) or MRI (Figure 2) marker significantly changed over the 2-year follow-up period These findings suggest that a substantially longer time interval than 2 years should be assessed to evaluate the Alzheimer's disease progression and biomarker-guided targeted therapies in presymptomatic SMCpos adults.

Background: Normal aging is accompanied by cognitive decline, structural and functional brain changes. Cognitive training is a potentially effective intervention for cognitive improvement. Transfer of training gains to untrained tasks is the ultimate goal of cognitive training. However, the neural mechanisms underlying successful transfer remain underinvestigated. Objective: To examine the predictive role of resting-state functional connectivity in the transfer of training gains. Methods: We analyzed resting-state fMRI and cognitive data of 181 healthy older adults (mean age: 68 years) who underwent a 4-week cognitive training at three study sites. The control group consisted of 54 older adults. Participants underwent neuropsychological assessments before and directly after the training, as well as 12 weeks after. We used aggregate scores representing working memory, memory and executive functions to assess transfer effects. Baseline resting-state fMRI was used to investigate functional connectivity. We used a seed-based and an independent component analysis approach to examine brain network activity. Results: The majority of our participants transferred cognitive training gains successfully over a three-month period. Baseline resting-state functional connectivity within the default mode network and the central executive network did not predict transfer of training gains. Conclusions: Baseline resting-state functional connectivity of large-scale networks does not appear to predict who will benefit from cognitive training in healthy older adults. These findings contribute to a better understanding of the functional brain mechanisms underlying transfer of training gains and highlight the need for larger, multi-modal neuroimaging studies to identify reliable neural predictors of cognitive training outcomes.

Inflammation is recognized as a key hallmark of Alzheimer’s disease (AD), alongside amyloid beta (Aβ) accumulation and tau pathology. Recent evidence suggests that neuroinflammatory markers in cerebrospinal fluid (CSF) are associated with progressive neurodegeneration and regional brain atrophy. In this study, we investigated the relationship between CSF inflammatory markers and atrophy in the basal forebrain and hippocampus longitudinally. We included 296 participants (37 with AD dementia, 69 with mild cognitive impairment (MCI), 98 with subjective cognitive decline (SCD) and 92 healthy control) from the DELCODE study. Data included baseline CSF markers, the Aβ42-phosphotau181 ratio, disease diagnosis, ApoE4 status, and longitudinal structural MRI volumes for specific brain regions, with a mean follow-up of 19.8 months (SD = 16.7). Latent factors were previously derived via Bayesian confirmatory factor analysis from 14 CSF markers and classified into Synaptic, Microglia, Chemokine/Cytokine, and Complement groups. We used linear mixed-effects models to assess interactions between latent factors and time on brain regions-controlling for age, sex, education and ApoE4 status- and, based on our systematic review, examined whether neurogranin, sTREM2, ferritin, and YKL40 predicted longitudinal changes. Our findings revealed significant interaction effects between specific biomarkers and regional brain atrophy. Longitudinal atrophy in the hippocampus was significantly associated with higher levels of the synaptic marker (β = -0.018, p = 0.004), sTREM2 (β = -0.012, p = 0.031), and YKL40 (β = -0.022, p = 0.0002). Similarly, increased levels of ferritin (β = -0.040, p = 0.024) and YKL40 (β = -0.041, p = 0.029) were predictive of longitudinal atrophy in the basal forebrain. In contrast, neurogranin and other latent factors—including microglia, chemokine/cytokine, and complement—did not show significant associations with atrophy in either brain region. These results suggest that certain biomarkers, particularly the synaptic latent factor and the individual markers sTREM2, ferritin, and YKL40, are predictive of longitudinal neurodegeneration in key brain regions vulnerable to Alzheimer's disease. The associations found with hippocampal and basal forebrain atrophy highlight the potential of these markers for tracking disease progression and improving early detection strategies.

The black-box nature of deep learning still prevents its widespread clinical use due to the high risk of hidden biases and prediction errors. Over the last decade, various explanation methods have been proposed to reveal the latent mechanisms of neural networks and support their decisions. However, interpreting the explanations themselves can be challenging, and there is still little consensus on how to evaluate the quality of explanations. To investigate the fidelity of explanations provided by prominent feature attribution methods for Convolutional Neural Networks in Alzheimer's Disease (AD) detection, this paper applies relevance-guided perturbation to the Magnetic Resonance Imaging (MRI) input images. According to the fidelity metric, the AD class probability showed the steepest decline when the perturbation was guided by Integrated Gradients or DeepLift. We conclude by highlighting the role of the reference image in feature attribution with regard to AD detection from MRI images. The source code for the experiments is publicly available on GitHub at https://github.com/bckrlab/ad-fidelity.

Background Imaging studies showed early atrophy of the cholinergic basal forebrain in prodromal sporadic Alzheimer's disease and reduced posterior basal forebrain functional connectivity in amyloid positive individuals with subjective cognitive decline. Similar investigations in familial cases of Alzheimer's disease are still lacking. Objectives To test whether presenilin-1 E280A mutation carriers have reduced basal forebrain functional connectivity and whether this is linked to amyloid pathology. Design This is a cross-sectional study that analyzes baseline functional imaging data. Setting We obtained data from the Colombia cohort Alzheimer's Prevention Initiative Autosomal-Dominant Alzheimer's Disease Trial. Participants We analyzed data from 215 asymptomatic subjects carrying the presenilin-1 E280A mutation [64% female; 147 carriers (M = 35 years), 68 noncarriers (M = 40 years)]. Measurements We extracted functional magnetic resonance imaging data using seed-based connectivity analysis to examine the anterior and posterior subdivisions of the basal forebrain. Subsequently, we performed a Bayesian Analysis of Covariance to assess the impact of carrier status on functional connectivity in relation to amyloid positivity. For comparison, we also investigated hippocampus connectivity. Results We found no effect of carrier status on anterior (Bayesian Factor10 = 1.167) and posterior basal forebrain connectivity (Bayesian Factor10 = 0.033). In carriers, we found no association of amyloid positivity with basal forebrain connectivity. Conclusions We falsified the hypothesis of basal forebrain connectivity reduction in preclinical mutation carriers with amyloid pathology. If replicated, these findings may not only confirm a discrepancy between familial and sporadic Alzheimer's disease, but also suggest new potential targets for future treatments.

Normal aging is associated with alterations of functional connectivity (FC) in brain neuronal networks. Altered network connectivity may be associated with accelerated cognitive decline. Physical activity is considered a beneficial lifestyle factor for maintaining cognitive health. Higher intensities of physical activity may induce structural and functional changes in the brain, particularly in regions involved in cognitive functions, such as memory, attention and executive functions. However, the underlying neural mechanisms are not widely investigated. Our aim was to examine the association between resting-state FC of brain networks and baseline physical activity in healthy older adults. We analyzed baseline resting-state fMRI and baseline physical activity data of 149 healthy older adults (mean age: 68 years) from the AgeGain study. Physical activity was measured by using actigraphs worn for 7 days. Different intensities were measured, such as light, mean and moderate-to-vigorous activity (min/d). We used Independent Component Analysis (ICA) and seed-based approaches to examine brain network activity in the Default Mode Network (DMN), Salience Network (SAL), Central Executive Network (CEN), Visual Network (VN) for cognitive effects and Sensorimotor Network (SMN) for physical effects. We observed statistically significant associations between functional activation within SMN and light physical activity and spatially restricted effects for DMN and moderate-to-vigorous physical activity ( p <.01 uncorrected). In addition, we observed an overlap on frontal activation across DMN, SMN and SAL. Results of the seed-based analysis will be presented at the conference. Light to higher intensities of physical activity showed an association with higher functional activation of networks previously associated with cognitive decline and physical activity. This agrees with the notion that physical activity may be a protective factor against cognitive decline. Further research is needed to test the replicability of these results.

When amyotrophic lateral sclerosis (ALS), a TDP-43 proteinopathy, and progressive supranuclear palsy (PSP), a tauopathy, are associated with frontotemporal dementia (ALS-FTD or PSP-FTD), clinical differentiation can be challenging. There are no established imaging biomarkers to differentiate ALS-FTD from PSP-FTD. We evaluated the midsagittal midbrain area (MBA) and the midbrain-to-pons-(MB/P)-ratios in T1 MPRAGE MRI of 36 PSP cases (n = 14 PSP-FTD), 77 ALS cases (n = 10 ALS-FTD), and 72 healthy controls (HC). In ALS, both parameters were indistinguishable from HC. Patients with ALS-FTD had low MBA-values and MB/P-ratios not significantly different from cases of PSP. While ROC-analyses provided an excellent diagnostic accuracy of both parameters for differentiating PSP from HC (AUCMBA = 0.974) as well as PSP from ALS (AUCMBA = 0.982), midbrain morphometry provided poor diagnostic accuracy for distinguishing ALS-FTD from PSP-FTD (AUCMBA = 0,614). The MBA and the MB/P-ratio are morphometric parameters that have proven reliable in atypical Parkinsonian syndromes. Both can distinguish between PSP and ALS in their typical clinical forms. However, they cannot differentiate between PSP-FTD and ALS-FTD.

Recent research has shown that cognitive reserve is associated with better cognitive abilities in ALS/MND, and that a slow brain ageing speed is associated with intact cognition in ALS. This study compares the effects of cognitive reserve and the predicted brain age difference (PAD) on the risk of being diagnosed with ALS, the risk of having cognitive or behavioral impairment, or even fronto-temporal dementia, and on disease duration.Our results indicated that neither PAD nor cognitive reserve was associated with an increased risk of ALS, but that higher PAD was associated with an increased risk of cognitive impairments and FTD, as well as a shortened disease duration. Higher cognitive reserve on the other hand was associated with a lower risk of cognitive impairment and a longer disease duration.Brain age as a proxy of brain reserve influences disease progression and presentation more strongly than cognitive reserve.

Introduction: The primary goal of developing new clinical diagnostic solutions is to create value for healthcare. The rapid rise of artificial intelligence (AI)-based diagnostics has led to a surge in publications and, to a lesser extent, market-ready tools. Clinicians must now integrate these innovations to manage increasing data volumes, making it challenging to assess the added value of new tools in the diagnostic workflow. Methods: The INTERREG Baltic Sea Region project “Clinical Artificial Intelligence-Based Diagnostics (CAIDX)” developed a comprehensive blueprint guiding the process from identifying clinical needs to implementing certified AI products in diagnostics. The approach emphasizes systematic evaluation at each development stage and throughout the AI solution’s lifecycle, incorporating diverse stakeholder perspectives and a range of evaluation methodologies. Results: The CAIDX project produced the “Clinical AI-Pathway,” an end-to-end framework for integrating AI-based diagnostic tools. This framework provides methodologies and tools for systematic evaluation at all stages, ensuring alignment with clinical needs and rigorous assessment of value. Conclusions: Systematic, multi-perspective evaluation is crucial for successfully integrating AI diagnostics into clinical practice. The “Clinical AI-Pathway” framework offers a structured method for assessing and implementing AI solutions, supporting their value-driven adoption in healthcare. The framework, available at ClinicalAI.eu, aims to facilitate broader and more effective use of AI in clinical diagnostics.

Explainable Artificial Intelligence (XAI) methods enhance the diagnostic efficiency of clinical decision support systems by making the predictions of a convolutional neural network’s (CNN) on brain imaging more transparent and trustworthy. However, their clinical adoption is limited due to limited validation of the explanation quality. Our study introduces a framework that evaluates XAI methods by integrating neuroanatomical morphological features with CNN-generated relevance maps for disease classification. We trained a CNN using brain MRI scans from six cohorts: ADNI, AIBL, DELCODE, DESCRIBE, EDSD, and NIFD (N = 3253), including participants that were cognitively normal, with amnestic mild cognitive impairment, dementia due to Alzheimer’s disease and frontotemporal dementia. Clustering analysis benchmarked different explanation space configurations by using morphological features as proxy-ground truth. We implemented three post-hoc explanations methods: (i) by simplifying model decisions, (ii) explanation-by-example, and (iii) textual explanations. A qualitative evaluation by clinicians (N = 6) was performed to assess their clinical validity. Clustering performance improved in morphology enriched explanation spaces, improving both homogeneity and completeness of the clusters. Post hoc explanations by model simplification largely delineated converters and stable participants, while explanation-by-example presented possible cognition trajectories. Textual explanations gave rule-based summarization of pathological findings. Clinicians’ qualitative evaluation highlighted challenges and opportunities of XAI for different clinical applications. Our study refines XAI explanation spaces and applies various approaches for generating explanations. Within the context of AI-based decision support system in dementia research we found the explanations methods to be promising towards enhancing diagnostic efficiency, backed up by the clinical assessments.

BackgroundThe early detection of Alzheimer's disease (AD) requires an understanding of the relationships between a wide range of features. Conditional independencies and partial correlations are suitable measures for these relationships, because they can identify the effects of confounding and mediating variables.ObjectiveTo estimate conditional dependencies and partial correlations between relevant features in AD using a Bayesian approach to Gaussian copula graphical models (GCGMs). This approach has two key advantages. First, it includes binary, discrete, and continuous variables. Second, it quantifies the uncertainty of the estimates. Despite these advantages, Bayesian GCGMs have not been applied to AD research yet.MethodsWe design a GCGM to find the conditional dependencies and partial correlations among brain-region specific gray matter volume and glucose uptake, amyloid-beta levels, demographic information, and cognitive test scores. We applied our model to 1022 participants, including healthy and cognitively impaired, across different stages of AD.ResultsWe found that aging reduces cognition through three indirect pathways: hippocampal volume loss, posterior cingulate cortex (PCC) volume loss, and amyloid-beta accumulation. We found a positive partial correlation between being woman and cognition, but also discovered four indirect pathways that dampen this association in women: lower hippocampal volume, lower PCC volume, more amyloid-beta accumulation, and less education. We found limited relations between brain-region specific glucose uptake and cognition, but discovered that the hippocampus and PCC volumes are related to cognition.ConclusionsThis study shows that the use of GCGMs offers valuable insights into AD pathogenesis.

Robust analysis of biosignals hinges on well-crafted processing pipelines, yet assembling them is still a slow, errorprone exercise, requiring both domain expertise and programming skills. While Large Language Models (LLMs) offer promising assistance, they fundamentally lack the combinatorial reasoning capabilities needed for designing reliable and reproducible processing pipelines. We present an innovative hybrid system that harnesses LLMs' natural language understanding while leveraging classical AI planning for logical reasoning and validation. A retrieval-augmented model parses natural language pipeline descriptions into implementation-independent atomic biosignal processing operations, which follow to a Hierarchical Task Network (HTN) domain that can plan and validate processing workflows, as well as flag violations of biosignal processing best practices. Evaluation on a preliminary corpus of scenarios demonstrates $\mathbf{0. 8 9 - 0. 9 8}$ precision in mapping natural language to processing blocks and 0.97-0.98 F1-score in generating valid 2-15 step pipelines using SHOP3 planner. A drag-and-drop pipeline design interface allows users to build and formally validate their pipeline ideas from scratch, showing potential to democratize and expedite biosignal analysis for scientists of different backgrounds.

Background Convolutional neural network (CNN) based volumetry of MRI data can help differentiate Alzheimer's disease (AD) and the behavioral variant of frontotemporal dementia (bvFTD) as causes of cognitive decline and dementia. However, existing CNN-based MRI volumetry tools lack a structured hierarchical representation of brain anatomy, which would allow for aggregating regional pathological information and automated computational inference. Objective Develop a computational ontology pipeline for quantifying hierarchical pathological abnormalities and visualize summary charts for brain atrophy findings, aiding differential diagnosis. Methods Using FastSurfer, we segmented brain regions and measured volume and cortical thickness from MRI scans pooled across multiple cohorts (N = 3433; ADNI, AIBL, DELCODE, DESCRIBE, EDSD, and NIFD), including healthy controls, prodromal and clinical AD cases, and bvFTD cases. Employing the Web Ontology Language (OWL), we built a semantic model encoding hierarchical anatomical information. Additionally, we created summary visualizations based on sunburst plots for visual inspection of the information stored in the ontology. Results Our computational framework dynamically estimated and aggregated regional pathological deviations across different levels of neuroanatomy abstraction. The disease similarity index derived from the volumetric and cortical thickness deviations achieved an AUC of 0.88 for separating AD and bvFTD, which was also reflected by distinct atrophy profile visualizations. Conclusions The proposed automated pipeline facilitates visual comparison of atrophy profiles across various disease types and stages. It provides a generalizable computational framework for summarizing pathologic findings, potentially enhancing the physicians’ ability to evaluate brain pathologies robustly and interpretably.

BackgroundSpeech-based features extracted from telephone-based cognitive tasks show promise for detecting cognitive decline in prodromal and manifest dementia. Little is known about the cerebral underpinnings of these speech features.ObjectiveTo examine associations between speech features, brain atrophy, and longitudinal cognitive decline in individuals at risk for Alzheimer's disease (AD).MethodsHealthy volunteers, individuals with subjective cognitive decline, and those with mild cognitive impairment completed phonebot-guided semantic verbal fluency (SVF) and 15-word verbal learning task (VLT). Speech features were automatically extracted, and a global cognitive score (SB-C score) was computed. We analyzed data from 161 participants for cognitive trajectories, 141 for cross-sectional brain atrophy, and 102 for longitudinal brain changes. Analyses were conducted using multiple linear regressions, mixed-effects models, and voxel-based morphometry.ResultsThe SB-C score was associated with bilateral hippocampal volumes, SVF features were primarily associated with left hemisphere regions, including the inferior frontal, parahippocampal, and superior/middle temporal gyri (puncorr < 0.001). SB-C score, SVF correct counts, and VLT delayed recall were associated with atrophy rates in the hippocampal/parahippocampal gyrus and left middle/inferior temporal gyri (pFDR < 0.05). These features were also associated with cognitive decline assessed via Preclinical Alzheimer's Cognitive Composite 5, SVF, and Wordlist learning delayed recall (pFDR < 0.01). Word frequency and temporal cluster switches showed varying associations with cognitive trajectories. Other features did not show robust associations.ConclusionsIn this study, we highlight the potential of digital speech features for identifying brain atrophy and cognitive decline over time in at-risk AD populations.

Speech features extracted from automated remote cognitive assessments correlate with performance on traditional cognitive tasks in individuals at risk of Alzheimer's Disease (AD), demonstrating their potential to support early diagnosis. However, the capability of these features to signal early AD-related brain changes remains less explored. Within the PROSPECT-AD study, 234 participants ranging from cognitively normal to mild cognitive impairment were recruited from the German DZNE longitudinal cohorts DELCODE and DESCRIBE. At home, all participants completed the phone-based and chatbot-guided Semantic Verbal Fluency task (SVF) and the Rey Auditory Verbal Learning Test (RAVLT). Linguistic and acoustic features were automatically extracted from phone call recordings using an AI model to calculate task-specific and composite cognitive scores. Structural MRI, functional MRI, and various paper-and-pencil cognitive scores were collected during cohort visits. We employed multiple linear regression, mixed-effects models, and independent component analysis (ICA), followed by voxel-wise post hoc analyses, to assess associations between digital speech-based indicators and: (1) cross-sectional brain atrophy ( n  = 108), (2) longitudinal brain atrophy ( n  = 90), (3) cross-sectional resting-state functional connectivity ( n  = 86), and additionally (4) trajectories of cognitive decline ( n  = 146). SVF correct counts were positively associated with brain volumes in the left temporal pole, left inferior, middle, and superior temporal gyri (The t (100) values ranged from 4.48 to 4.96) in voxel-wise analyses (Figure 1a). Longitudinal analyses indicated that higher SVF correct counts were linked to slower rates of hippocampal and anterior cingulate atrophy (Figure 1b). Functional connectivity analyses suggested that SVF features, such as word frequency, were associated with rsFC areas within the default mode network (The t (71) values ranged from 3.65 to 3.85) (Figure 1c). Higher composite cognitive scores, along with SVF and RAVLT features, were associated with slower cognitive decline, as measured by established paper-and-pencil cognitive assessments, including the Preclinical Alzheimer Cognitive Composite (PACC) 5, SVF, and RAVLT delayed recall (Figure 2). Phone-based cognitive assessments hold promise as a remote and scalable tool for identifying AD-related structural and functional brain changes. They offer predictive value for cognitive trajectories in pre-dementia populations. This approach could aid in identifying individuals at risk while guiding further evaluation, broadening their utility beyond cognitive screening.