Artificial Intelligence and Digital Neuropsychology: Opportunities and Challenges of Smartphone-Based Cognitive Assessment, Wearables, Digital Biomarkers, and Machine Learning
The rapid development of artificial intelligence (AI), mobile health technologies, and wearable sensing systems is transforming the field of clinical neuropsychology. Traditional neuropsychological assessment relies predominantly on standardized cognitive tests administered under controlled clinical conditions. Although these methods provide valuable diagnostic information, they have limitations concerning ecological validity, assessment frequency, accessibility, and sensitivity to subtle cognitive changes.
Digital neuropsychology offers new opportunities to overcome some of these limitations through smartphone-based cognitive assessments, wearable sensors, digital biomarkers, and machine learning algorithms. These technologies facilitate continuous or repeated monitoring of cognitive performance and everyday functioning, potentially improving the early identification of neurodegenerative diseases, longitudinal monitoring, and personalized clinical decision-making. Nevertheless, their implementation raises considerable methodological, psychometric, ethical, and regulatory challenges.
Particular concerns include the validity and reliability of digitally derived measures, algorithmic bias, data privacy, technological inequalities, and the interpretability of machine learning predictions. This narrative review examines the theoretical foundations, current applications, and clinical potential of AI-supported digital neuropsychology, with particular emphasis on smartphone-based assessment, wearable technologies, digital biomarkers, and machine learning. It critically evaluates the opportunities and limitations of these approaches and identifies priorities for research and clinical implementation. The review argues that digital technologies should be understood primarily as complementary instruments that can extend conventional neuropsychological assessment rather than replace comprehensive clinical evaluation. Their responsible integration requires rigorous psychometric validation, interdisciplinary collaboration, transparent algorithmic development, and person-centered ethical governance.
Neuropsychology is undergoing a fundamental technological transformation driven by advances in artificial intelligence, mobile computing, wearable sensors, and digital health infrastructures. Conventional neuropsychological assessment has historically depended on standardized paper-and-pencil tests, structured clinical interviews, behavioral observations, and increasingly computerized testing procedures. These methods remain essential for evaluating cognitive domains such as memory, attention, executive functioning, processing speed, language, and visuospatial abilities. However, conventional assessment frequently captures cognitive performance at a limited number of time points and under highly standardized conditions that may not adequately reflect the complexity of everyday cognitive functioning.
Harris et al. (2024) emphasized that traditional neuropsychological testing is constrained by relatively infrequent measurement, limited ecological validity, and practical barriers to repeated assessment. In particular, cognitive performance may fluctuate according to sleep quality, fatigue, emotional state, environmental demands, and neurological symptoms. Single-session assessments may consequently provide an incomplete representation of an individual's functioning.
The increasing availability of smartphones and wearable technologies offers opportunities to investigate cognitive functioning beyond the clinical examination room. Smartphones can administer brief cognitive tests repeatedly, while wearable sensors capture behavioral and physiological information during everyday activities. Combined with machine learning methods, these technologies may identify patterns of cognitive and functional change that remain undetected through conventional testing.
The concept of digital phenotyping, introduced into contemporary mental health research by Onnela and Rauch (2016), provides an important theoretical foundation for this development. Digital phenotyping involves the quantification of human behavior and functioning through information collected by personal digital devices. Such data may include active measurements, such as cognitive task responses, and passive measurements, such as movement patterns, device interactions, and sleep-related signals.
Within neuropsychology, these approaches potentially enable more continuous, individualized, and context-sensitive assessment strategies. Of particular interest is their application to mild cognitive impairment (MCI), Alzheimer's disease, Parkinson's disease, multiple sclerosis, traumatic brain injury, and other neurological conditions characterized by cognitive or behavioral changes.
Kourtis et al. (2019) proposed that mobile and wearable technologies could contribute to the identification of early cognitive, behavioral, and motor changes associated with Alzheimer's disease. Nevertheless, the transition from experimental digital technologies to clinically established neuropsychological instruments remains challenging.
The central objective of this article is therefore to critically examine how artificial intelligence and digital technologies may transform neuropsychological assessment, while identifying the scientific, methodological, and ethical requirements necessary for their responsible clinical application.
Conceptual Foundations of Digital Neuropsychology
Definition and scope
Digital neuropsychology can be understood as the application of digital technologies to the measurement, interpretation, and monitoring of cognitive, emotional, and behavioral functioning. It includes computerized neuropsychological assessment, mobile cognitive testing, ecological momentary assessment, remote monitoring, wearable sensing, digital biomarker development, and AI-assisted analysis.
Importantly, digital neuropsychology extends beyond simply transferring established paper-based tests onto electronic devices. It creates opportunities to develop fundamentally different assessment methods that incorporate temporal variability, everyday environmental conditions, and multidimensional behavioral information.
Harris et al. (2024) argued that contemporary digital technologies can expand neuropsychological practice through smartphone sensors, ecological momentary assessment, wearable devices, speech analysis, and passive behavioral monitoring. These methods allow researchers to investigate cognitive functioning as a dynamic process rather than exclusively as a stable characteristic measured during a clinical appointment.
This conceptual shift has important implications for neuropsychological theory. Conventional psychometric models frequently emphasize differences between individuals, whereas digital longitudinal methods also permit investigation of fluctuations within the same individual over time.
For example, rather than determining whether an individual's memory performance is lower than that of an age-matched reference population at a single time point, digital assessment may determine whether their recent memory performance deviates significantly from their own established baseline.
Ecological validity and intensive longitudinal assessment
Ecological validity refers to the extent to which an assessment reflects or predicts functioning in everyday environments. Conventional neuropsychological tests generally use controlled conditions to maximize standardization and internal validity. However, the complexity of everyday activities may not be fully represented by such conditions.
Smartphone-based assessments can address aspects of this limitation by measuring cognition in natural environments. Participants may complete brief tasks during normal daily activities, allowing cognitive performance to be examined in relation to contextual factors such as fatigue, stress, time of day, or physical activity.
Ecological momentary assessment (EMA) provides a methodological framework for repeated real-time measurements of psychological and behavioral states. Harris et al. (2024) described how EMA can reduce retrospective reporting bias and reveal short-term variations in cognitive functioning that conventional assessments may overlook.
Nevertheless, conducting tests outside controlled clinical environments introduces additional sources of measurement variability. Environmental distractions, interruptions, differences in device characteristics, and inconsistent participant engagement may influence test scores.
Consequently, improved ecological relevance does not automatically imply superior psychometric validity. The potential benefits of naturalistic assessment must be balanced against the need for standardized measurement conditions and rigorous interpretation.
Artificial intelligence as an analytical framework
Artificial intelligence encompasses computational approaches designed to perform tasks involving pattern recognition, prediction, classification, and decision support. Within digital neuropsychology, AI is primarily applied through machine learning, including supervised learning, unsupervised learning, and deep learning.
These methods can analyze complex and high-dimensional datasets generated by smartphones, wearable sensors, and computerized cognitive tests.
Traditional statistical analyses generally examine prespecified relationships among variables. Machine learning approaches can additionally identify nonlinear relationships, interactions, and multivariate patterns that may be difficult to recognize through simpler analytical methods.
However, the distinction between statistical modeling and machine learning should not be overstated. Both require appropriate research designs, representative datasets, careful validation, and clinically meaningful interpretation.
The primary contribution of AI is therefore not that it eliminates uncertainty, but that it offers additional methods for analyzing complex longitudinal and multimodal information.
Smartphone-Based Cognitive Assessment
Opportunities for repeated and accessible cognitive testing
Smartphones represent an important technological platform for digital neuropsychological assessment because they combine computing capabilities, touch-sensitive interfaces, microphones, motion sensors, and wireless connectivity in portable devices.
Smartphone-based cognitive assessment typically involves short, digitally administered tasks designed to measure particular cognitive abilities.
Common assessment domains include:
Memory: Immediate recall, delayed recognition, associative learning, and working memory.
Attention: Sustained attention, selective attention, and response inhibition.
Executive functioning: Cognitive flexibility, inhibitory control, planning, and task switching.
Processing speed: Reaction time, response accuracy, and decision speed.
Language: Verbal fluency, naming, and speech production.
Visuospatial functioning: Spatial memory, visual search, and perceptual discrimination.
Repeated administration can generate longitudinal cognitive profiles that may be more informative than isolated measurements.
Harris et al. (2024) highlighted the potential of smartphone-based methods to extend cognitive assessment into everyday life and increase the frequency of data collection.
For individuals with progressive neurological conditions, repeated assessments may support the detection of gradual changes in cognitive functioning. In disorders characterized by fluctuating symptoms, such as multiple sclerosis or certain neurodegenerative conditions, repeated testing may also capture variations associated with fatigue and disease activity.
Another important advantage concerns accessibility. Remote administration may reduce travel requirements and provide additional opportunities for individuals with mobility limitations or restricted access to specialist neuropsychological services.
Nevertheless, greater technological availability does not necessarily guarantee equitable access. Smartphone ownership, digital literacy, visual or motor impairments, and familiarity with touchscreen interfaces may substantially influence participation and performance.
Measurement precision and digital behavioral indicators
Digital cognitive tests offer opportunities to collect information beyond conventional accuracy-based scores.
For example, touchscreen interactions can provide detailed measurements of response latency, movement trajectories, hesitation, error correction, and variability across repeated responses.
These behavioral characteristics may contain information about cognitive processes that cannot easily be observed through traditional scoring procedures.
Digital testing also facilitates automated stimulus presentation, standardized instructions, electronic recording, and calculation of multiple performance indicators.
However, technical precision should not be confused with psychological validity. Although a smartphone may record responses with considerable temporal resolution, the clinical meaning of small differences in response latency depends on the reliability of the hardware, software, testing conditions, and cognitive construct being measured.
Harris et al. (2024) emphasized that digital assessment instruments require appropriate validation and normative data before they can be interpreted reliably in clinical populations.
Individualized cognitive trajectories
A particularly promising application involves the development of individualized cognitive reference profiles.
Traditional normative interpretation compares an individual's performance against distributions derived from representative populations. Digital longitudinal assessment can supplement this approach by establishing person-specific baselines.
For example, an individual may demonstrate cognitive performance within conventional normative limits while exhibiting a persistent decline relative to their own previous functioning.
Longitudinal digital measurements may help identify such changes, provided that statistical models appropriately account for measurement error, practice effects, and normal day-to-day variability.
An individualized assessment approach could therefore be useful for monitoring disease progression, evaluating rehabilitation outcomes, and investigating treatment-related cognitive changes.
Nevertheless, substantial research is required to determine the magnitude of change that should be considered clinically meaningful.
Psychometric and methodological limitations
The validity of smartphone-based cognitive assessment cannot be assumed merely because a digital task resembles an established neuropsychological test.
Several methodological considerations are particularly important.
First, construct validity must be demonstrated. A task intended to measure working memory, for example, should provide evidence that performance primarily reflects working memory rather than motor speed, technological familiarity, or other competing influences.
Second, test–retest reliability must be evaluated, especially when measurements are intended to detect subtle changes over time.
Third, repeated administration may produce practice effects. Improvements in performance may reflect increasing familiarity with the task rather than genuine cognitive improvement.
Fourth, variations in device hardware and software can affect stimulus presentation and response recording.
Finally, remote testing introduces differences in environmental conditions, participant motivation, and external assistance.
These limitations indicate that digital cognitive assessments require dedicated psychometric research, including examination of measurement invariance, reliable change indices, normative reference values, and longitudinal sensitivity.
The clinical objective should not simply be to maximize the quantity of collected data, but to ensure that digitally measured cognitive indicators are scientifically interpretable and clinically useful.
Wearable Technologies and Continuous Cognitive Monitoring
Physiological and behavioral sensing
Wearable technologies represent another important component of digital neuropsychology. Smartwatches, activity trackers, smart rings, and other sensor-based devices can collect longitudinal measurements of physiological and behavioral activity with relatively limited participant involvement.
Depending on the device, these measurements may include movement, gait characteristics, physical activity, heart rate, heart rate variability, sleep-related activity patterns, and electrodermal activity.
Although these variables do not directly measure cognition, they may provide clinically relevant contextual information about neurological and psychological functioning.
Kourtis et al. (2019) discussed the potential of wearable technologies to identify behavioral and physiological changes associated with Alzheimer's disease, including alterations in movement, sleep patterns, and autonomic nervous system functioning.
For example, changes in walking speed, gait variability, or motor coordination may occur alongside cognitive decline in certain neurological conditions. Similarly, disrupted sleep–wake cycles may influence attention, memory consolidation, and executive functioning.
Wearable technologies may therefore contribute to a more comprehensive understanding of the relationship between cognitive functioning and everyday behavioral patterns.
However, these measures are generally nonspecific. Reduced physical activity, for instance, may reflect neurological deterioration, depressive symptoms, chronic pain, medication effects, environmental restrictions, or changes in lifestyle.
Consequently, wearable-derived measures require interpretation within a broader clinical and contextual framework.
Passive monitoring and digital phenotyping
Passive monitoring involves the collection of data through sensors or device interactions without requiring participants to complete an explicit assessment task each time.
In contrast to active assessment, which depends on deliberate engagement, passive sensing can generate repeated or continuous measurements during everyday activities.
Sameh et al. (2025) conducted a systematic review of machine learning approaches using passive, noninvasive signals collected through wearable devices and smartphones. The review included 66 studies and identified applications across several health domains, including fatigue, stress, depression, seizures, and Parkinson's disease.
The authors highlighted the potential of passive sensing for health monitoring and predictive modeling, while emphasizing unresolved challenges associated with data quality, missing observations, and reproducibility.
For digital neuropsychology, passive monitoring may be particularly valuable when patients experience difficulties completing repeated cognitive tasks.
Nevertheless, passive data collection raises significant questions regarding consent, transparency, privacy, and the distinction between clinically relevant monitoring and excessive surveillance.
Furthermore, the availability of large quantities of passively collected data does not guarantee their clinical relevance. Measurements must be linked to clearly defined neuropsychological constructs and meaningful clinical outcomes.
Sleep, motor functioning, and cognitive health
Sleep and motor functioning are especially relevant to neuropsychological research because they are closely associated with cognitive performance and neurological disease.
Sleep disturbances may affect sustained attention, working memory, emotional regulation, and executive functioning. Wearable devices may provide estimates of sleep duration, sleep timing, and rest–activity rhythms across extended observation periods.
Similarly, wearable movement sensors may capture walking patterns, physical activity, and motor variability.
Kourtis et al. (2019) identified such sensor-derived characteristics as potential sources of information for the study of early Alzheimer's disease.
More recent evidence further illustrates both the promise and limitations of this research area. Gramkow et al. (2025) reviewed 109 studies investigating wearable- or portable-device-derived digital biomarkers in early Alzheimer's disease. The most frequently studied domains included rest–activity patterns, speech, and graphomotor functioning. However, relatively few studies evaluated the diagnostic or prognostic utility of these measures.
This distinction is important. Demonstrating an association between a sensor-derived variable and cognitive impairment is not equivalent to establishing that the variable can diagnose disease, predict clinically relevant outcomes, or guide treatment decisions.
Future studies must therefore demonstrate whether wearable-derived information provides incremental diagnostic or prognostic value beyond existing clinical assessments.
Digital Biomarkers in Neuropsychological Assessment
Definition and clinical relevance
Digital biomarkers are quantifiable indicators derived from digital technologies that are intended to reflect biological, physiological, behavioral, or health-related processes.
Within digital neuropsychology, candidate digital biomarkers may be derived from cognitive task performance, speech characteristics, movement patterns, touchscreen interactions, or combinations of these measures.
A fundamental distinction must be made between a digital measurement and a clinically validated biomarker.
For example, a smartphone application may measure response times during a memory task. However, response time becomes a clinically meaningful digital biomarker only when sufficient evidence supports its reliability, validity, and relevance to a specified clinical purpose.
The U.S. Food and Drug Administration (FDA, 2023) emphasized the importance of evaluating whether digital health technologies are appropriate for their intended use, particularly when employed in clinical investigations.
This principle is central to digital neuropsychology because the validity of a digital measure depends not only on technical measurement accuracy but also on its clinical interpretation and context of use.
Cognitive and behavioral digital biomarkers
Potential neuropsychological digital biomarkers can be classified into several broad domains.
Domain | Example digital measurements | Potential clinical relevance |
Cognitive performance | Response accuracy, reaction time, intraindividual variability | Monitoring attention, memory, and executive functioning |
Speech and language | Speech rate, pauses, lexical diversity, verbal fluency | Investigation of language-related and cognitive changes |
Motor functioning | Gait speed, movement variability, fine motor coordination | Monitoring neurological and functional changes |
Sleep and circadian activity | Sleep timing, rest–activity rhythms, sleep-duration estimates | Contextualizing cognitive fluctuations and daily functioning |
Smartphone interactions | Typing rhythm, touchscreen movements, task completion patterns | Characterizing behavioral changes and processing efficiency |
Multimodal behavioral profiles | Combined cognitive, physiological, and sensor-derived measures | Investigating complex longitudinal patterns and possible risk indicators |
Note. These are candidate measurements and possible applications, not universally validated diagnostic biomarkers. Their clinical utility depends on the population, measurement procedure, and intended purpose.
Speech analysis provides an illustrative example. Changes in word retrieval, speech fluency, hesitation, and linguistic complexity may accompany neurological conditions affecting language and cognition.
AI-supported speech analysis can quantify temporal and linguistic features that are difficult to evaluate consistently through unaided clinical observation.
Likewise, digital graphomotor measurements may characterize fine motor functioning and cognitive-motor integration through drawing or writing tasks completed on touchscreen devices.
Gramkow et al. (2025) demonstrated that speech, activity-related measures, and graphomotor characteristics are prominent areas of digital biomarker research in early Alzheimer's disease.
Nevertheless, the same observable characteristic may have multiple explanations. Reduced verbal fluency, for example, can be associated with cognitive impairment, anxiety, depression, language proficiency, fatigue, or differences in educational background.
Such indicators should therefore not be interpreted as disease-specific without appropriate validation.
Early detection of neurodegenerative diseases
One of the most promising applications of digital biomarkers is the detection of subtle functional changes that may precede clinically evident impairment.
Alzheimer's disease is of particular interest because cognitive and behavioral changes may emerge gradually over extended periods.
Kourtis et al. (2019) suggested that mobile and wearable technologies could help characterize early changes across cognitive, motor, behavioral, and physiological domains.
Digital assessments may be especially informative when they capture longitudinal trajectories rather than isolated abnormalities.
For instance, progressively increasing response variability, declining task accuracy, or changes in habitual activity patterns might suggest the need for further clinical evaluation.
However, the distinction between detecting cognitive change and diagnosing a particular neurodegenerative disease must be maintained.
The revised Alzheimer's Association diagnostic framework emphasizes the biological characterization of Alzheimer's disease and the role of established disease-related biomarkers (Jack et al., 2024). Smartphone-derived behavioral indicators and wearable measurements generally do not provide direct evidence of Alzheimer's-specific neuropathology.
Consequently, digital neuropsychological biomarkers should primarily be viewed as potential measures of cognitive or functional manifestations, risk stratification, or longitudinal monitoring unless stronger disease-specific validation becomes available.
Clinical validity and clinical utility
The successful development of a digital biomarker requires several distinct forms of evidence.
Analytical validation establishes whether the technology measures its intended signal accurately and reliably.
Clinical validation determines whether the measurement is meaningfully associated with the clinical condition or outcome of interest.
Clinical utility concerns whether the measurement improves decision-making or patient outcomes in a real-world clinical context.
These dimensions should not be treated as interchangeable.
A digital biomarker may exhibit excellent technical reliability while demonstrating limited diagnostic specificity. Similarly, a machine learning model may accurately classify participants in a research dataset without producing meaningful improvements in clinical care.
Gramkow et al. (2025) identified an important gap between the expanding research literature on digital biomarkers and evidence supporting their diagnostic, prognostic, or treatment-monitoring applications.
This suggests that future research should prioritize clinically meaningful outcomes, longitudinal validation, and evaluation of incremental value rather than exclusively developing additional measurement indicators.
Machine Learning and Artificial Intelligence in Digital Neuropsychology
Machine learning approaches
Machine learning methods are increasingly used to analyze the high-dimensional datasets generated by digital cognitive assessments and wearable sensors.
In supervised learning, algorithms are trained using observations with known outcomes, such as diagnostic categories or cognitive scores.
Common approaches include logistic regression with regularization, support vector machines, random forests, gradient boosting, and neural networks.
Unsupervised learning methods, including clustering and dimensionality reduction, can identify patterns within datasets without requiring predefined diagnostic labels.
Deep learning approaches may be particularly useful for complex data such as speech recordings, movement trajectories, and longitudinal sensor signals.
Sameh et al. (2025) documented the use of diverse machine learning strategies in research involving passive smartphone and wearable data. Their findings illustrated the potential of computational methods to integrate behavioral information for health-related prediction.
Within digital neuropsychology, these methods may support three broad objectives: predicting clinical outcomes, identifying meaningful behavioral patterns, and developing individualized models of cognitive change.
Nevertheless, the complexity of an algorithm does not necessarily correspond to superior predictive performance or clinical relevance. Simpler, interpretable models may be preferable when they achieve comparable results.
Multimodal data integration
A major opportunity for AI involves combining heterogeneous data sources into integrated models.
Traditional neuropsychological assessment primarily relies on cognitive performance measures, self-report instruments, and clinical observations. Digital methods allow these sources to be supplemented with physiological and behavioral data.
For example, an integrated assessment system could combine repeated memory-task performance, sleep-related activity, walking speed, and speech characteristics.
Machine learning algorithms might then identify relationships among these variables that would be difficult to detect through isolated measurement.
Multimodal integration may be particularly relevant because cognitive impairment rarely manifests exclusively within a single behavioral domain.
However, additional data modalities can introduce measurement noise, missing observations, and complex confounding relationships.
An increase in the number of predictors may also elevate the risk of overfitting, especially when datasets contain relatively few participants but large numbers of repeated measurements.
Therefore, multimodal models must demonstrate that the inclusion of additional data sources improves predictive accuracy, calibration, generalizability, or clinical utility.
Prediction of cognitive decline
Predictive modeling represents one of the most frequently discussed applications of AI-supported digital neuropsychology.
Models may aim to estimate the probability of cognitive deterioration, identify individuals at elevated risk of impairment, or characterize future changes in functioning.
The potential advantages include earlier referral for comprehensive clinical assessment and improved monitoring of patients with known neurological conditions.
However, predictive performance must be evaluated carefully.
Accuracy alone is often insufficient, particularly when the prevalence of a target condition is low. Sensitivity, specificity, positive predictive value, negative predictive value, discrimination, and calibration may all be relevant, depending on the intended use.
Furthermore, models must be evaluated on data independent of the training sample.
A particularly important methodological concern arises when multiple observations from the same participant are distributed across training and test datasets. Such procedures can produce overly optimistic performance estimates because the algorithm may recognize person-specific characteristics rather than generalizable disease patterns.
Participant-level separation, external validation, and prospective evaluation are therefore essential.
Recent evidence demonstrates the importance of these requirements. Qi et al. (2025) reviewed 86 AI models involving Alzheimer's disease-related digital biomarkers and reported that only two studies included external validation and three evaluated model calibration. These findings highlight a substantial gap between promising experimental performance and the evidence required for reliable clinical implementation.
npj Digital Medicine
Explainability and interpretability
The implementation of AI in clinical neuropsychology also requires attention to algorithmic interpretability.
Some machine learning methods produce complex predictions that are difficult to explain in terms of recognizable cognitive or physiological processes.
This is problematic when model outputs may influence clinical referrals, diagnostic investigations, or treatment-related decisions.
Clinicians need to understand the evidential basis, reliability, and limitations of predictions.
Explainable AI methods may help identify variables contributing to a model's output. However, an explanation of the mathematical factors influencing a prediction does not necessarily establish a causal relationship between those factors and neurological disease.
For example, a model may identify reduced physical activity as an important predictor of cognitive impairment. This does not demonstrate that reduced activity causes cognitive deterioration or that the relationship is specific to neurodegenerative pathology.
Clinical interpretation therefore requires integration of algorithmic outputs with neurological history, neuropsychological examination, medical findings, and contextual information.
AI should support rather than substitute clinical reasoning.
Clinical Applications and Opportunities
Screening and early identification
Digital technologies may provide scalable opportunities for preliminary cognitive screening, particularly in populations with limited access to specialist neuropsychological services.
Short smartphone-based cognitive tasks could facilitate repeated screening and identify individuals who may benefit from comprehensive clinical evaluation.
However, screening systems must be designed to balance sensitivity and specificity in the intended population.
False-positive classifications may result in unnecessary anxiety, additional examinations, and increased healthcare utilization. Conversely, false-negative results may delay appropriate clinical assessment.
Therefore, digital screening should not be presented as equivalent to formal diagnosis.
Longitudinal disease monitoring
Repeated digital assessment may be especially valuable for monitoring progressive neurological conditions.
Longitudinal measurements can potentially reveal changes that are not apparent during widely spaced clinical appointments.
In Alzheimer's disease and related disorders, digital measurements may complement established cognitive scales and functional assessments.
In Parkinson's disease, combinations of motor and cognitive measures may support investigation of relationships between movement symptoms and cognitive functioning.
For multiple sclerosis, repeated assessments may provide information about fluctuations associated with fatigue, cognitive symptoms, and disease activity.
These potential applications remain dependent on population-specific validation and evidence that digital monitoring improves clinically relevant decisions.
Personalized neuropsychological rehabilitation
Digital assessment may also contribute to individualized rehabilitation.
Repeated cognitive measurements could help clinicians characterize changes in performance over time, examine associations with everyday activities, and adjust rehabilitation strategies according to patient needs.
For example, a patient undergoing cognitive rehabilitation might complete brief attention or working-memory tasks at regular intervals. These results could be evaluated alongside fatigue ratings and activity-related measurements.
Such information may assist in identifying conditions under which performance deteriorates or improves.
However, improvements in digital task performance do not necessarily indicate generalization to everyday functioning.
The effectiveness of digitally informed rehabilitation must therefore be evaluated using meaningful functional outcomes rather than relying exclusively on changes in app-based scores.
Decentralized clinical research
Digital technologies may facilitate decentralized research designs in which participants complete assessments outside conventional research facilities.
The FDA (2023) recognized the potential of remote digital data acquisition to improve research efficiency, participant convenience, and access to clinical investigations.
In neuropsychological research, this approach could support more frequent observations, geographically distributed recruitment, and investigations of cognitive trajectories over extended periods.
Nevertheless, decentralized methods require robust protocols for informed consent, participant safety, device suitability, data quality, and the management of missing observations.
These requirements are particularly important when participants have cognitive impairment that may affect their ability to understand instructions or use digital technologies independently.
Methodological, Ethical, and Regulatory Challenges
Reliability, validity, and standardization
Despite their potential advantages, digital neuropsychological technologies face substantial methodological challenges.
Traditional neuropsychological instruments have often undergone extensive standardization and psychometric evaluation. Comparable evidence is not consistently available for emerging smartphone applications, wearable-derived measures, or AI-supported assessment systems.
Harris et al. (2024) emphasized that digital assessment technologies must be critically examined with respect to reliability, validity, normative interpretation, and their relationship to established neuropsychological measures.
The validity of digital tests may also be affected by operating-system updates, differences in device hardware, and modifications to software algorithms.
Consequently, a digital instrument validated on one technological platform cannot automatically be assumed to perform equivalently on another.
Longitudinal studies must also address practice effects, measurement error, selective participant dropout, and changes in device use.
Standardization therefore requires both psychometric and technological quality assurance.
Algorithmic bias and population representativeness
Machine learning models are influenced by the characteristics of the datasets used for their development.
When training datasets disproportionately represent particular age groups, educational backgrounds, languages, socioeconomic groups, or technological experiences, predictive models may perform less reliably in underrepresented populations.
This concern is especially relevant to neuropsychology, where cognitive performance is influenced by demographic, cultural, educational, and linguistic factors.
For example, smartphone interaction patterns may differ considerably between individuals with extensive technological experience and those with limited digital literacy.
An algorithm that interprets slower touchscreen interactions as evidence of cognitive impairment may consequently produce misleading results if it does not adequately account for motor functioning, technological familiarity, and device characteristics.
Algorithmic fairness should therefore be investigated through subgroup-specific performance analyses, representative sampling strategies, and external validation across diverse clinical populations.
Importantly, statistical corrections alone cannot fully resolve inequalities caused by differences in technological access or participation.
Data privacy and informed consent
Digital neuropsychological systems may collect highly sensitive information concerning cognition, behavior, mobility, sleep, communication, and everyday routines.
Even when individual measurements appear relatively harmless, combinations of data may reveal substantial information about personal functioning and health.
Onnela and Rauch (2016) identified the potential of smartphone-based digital phenotyping while recognizing the importance of responsible approaches to personal digital information.
Privacy concerns become particularly significant when continuous monitoring generates data beyond what is strictly necessary for the intended clinical purpose.
Appropriate safeguards include data minimization, secure storage, controlled access, transparent governance, and clear communication about the purposes of data processing.
Where possible, privacy-preserving analytical methods should be considered, including approaches that reduce the transfer or retention of directly identifiable information.
Informed consent must explain not only what information will be collected, but also how it will be analyzed, who may access it, how long it will be retained, and whether it may be reused for additional research purposes.
For individuals with cognitive impairment, consent procedures may require additional support and periodic reassessment.
Autonomy and psychological consequences
Continuous cognitive monitoring may influence how individuals perceive their own health and cognitive abilities.
Repeated alerts concerning possible deterioration could increase anxiety or encourage excessive attention to normal performance fluctuations.
Furthermore, automated risk predictions may be misunderstood as definitive clinical diagnoses.
These concerns underline the importance of contextualized feedback, appropriate communication, and access to professional interpretation.
Digital neuropsychological assessment should promote patient autonomy and informed participation rather than create dependency on algorithmic judgments.
The degree and frequency of monitoring should be proportionate to the intended clinical benefit.
Regulatory considerations
The clinical implementation of digital neuropsychological systems requires attention to applicable medical-device, data-protection, and research regulations.
The FDA (2023) issued guidance concerning digital health technologies used for remote data acquisition in clinical investigations, emphasizing the suitability of technologies for their intended purposes.
Regulatory expectations may differ depending on whether a digital application provides general wellness information, supports clinical research, assists professional decision-making, or performs a regulated medical-device function.
AI-supported systems introduce additional challenges because predictive performance may change following algorithm updates, modifications to data processing, or shifts in the populations using the technology.
Appropriate governance therefore requires ongoing monitoring of model performance, documentation of changes, and mechanisms for addressing clinically significant failures.
Regulatory compliance should be considered a necessary component of responsible implementation, although compliance alone does not establish clinical effectiveness.
Future Research Directions
The development of digital neuropsychology requires a transition from technology-driven experimentation toward clinically oriented evidence generation.
Several priorities are particularly important.
Large-scale longitudinal validation
Future research should prioritize prospective longitudinal studies involving diverse clinical and nonclinical populations.
Such studies can clarify whether digital measures reliably distinguish normal cognitive variability from clinically meaningful changes.
They should also determine whether digital measures predict functional deterioration or diagnostic outcomes beyond information obtained through conventional assessment.
Integration of multimodal information
Combining cognitive, behavioral, physiological, and clinical information may improve understanding of neurological functioning.
However, multimodal AI models should be evaluated against simpler alternatives to establish whether additional data actually improve clinical performance.
Model development should prioritize transparent validation, appropriate calibration, and generalizability rather than algorithmic complexity alone.
Individualized assessment models
Person-specific longitudinal models may represent an important future direction.
Such systems could evaluate deviations from established individual baselines while accounting for expected fluctuations, practice effects, and contextual influences.
Nevertheless, individualized models require sufficient observations, robust statistical procedures, and validated thresholds for clinically significant change.
Human-centered technology development
Patients, caregivers, neuropsychologists, neurologists, software developers, and researchers should be involved in the design and evaluation of digital assessment systems.
Participatory development may improve usability, accessibility, and acceptability, especially among older adults and individuals with neurological impairments.
Digital technologies must accommodate differences in sensory, motor, cognitive, and linguistic abilities.
Clinical implementation studies
A central limitation of the existing evidence is the gap between technological feasibility and demonstrated clinical benefit.
Future studies should evaluate whether digital neuropsychological assessment improves diagnostic pathways, treatment monitoring, rehabilitation planning, patient functioning, or healthcare accessibility.
Randomized or appropriately controlled implementation studies may be particularly valuable when comparing digitally enhanced assessment pathways with standard care.
Economic evaluations are also necessary to determine whether the benefits of remote monitoring justify the costs of devices, software, clinical oversight, data management, and technical support.
Transparent and reproducible AI research
Research transparency is essential for establishing confidence in AI-supported neuropsychological systems.
Sameh et al. (2025) highlighted the need for improved reporting of datasets, missing information, analytical procedures, and machine learning implementations.
Future research should therefore document preprocessing, feature selection, model development, validation procedures, calibration, and uncertainty.
External replication and independent evaluation should be prioritized before high-stakes clinical use.
Discussion
Artificial intelligence and digital technologies offer substantial opportunities to expand the methodological foundations of neuropsychology.
Their most significant contribution may be the transition from relatively infrequent, clinic-based assessments toward repeated, context-sensitive, and multidimensional measurements of cognitive functioning.
Smartphone-based cognitive assessments can support longitudinal characterization of memory, attention, executive functioning, and processing speed. Wearable technologies can supplement these measures by capturing activity, movement, sleep-related patterns, and physiological characteristics. Digital biomarkers may transform complex behavioral signals into quantifiable indicators, while machine learning provides computational methods for analyzing longitudinal and multimodal datasets.
Nevertheless, the evidence supporting these technologies remains heterogeneous.
Harris et al. (2024) described substantial opportunities for digital approaches to address limitations of conventional neuropsychological practice. Similarly, Kourtis et al. (2019) identified the potential of mobile and wearable technologies for investigating early changes associated with Alzheimer's disease.
More recent reviews illustrate that the field has advanced considerably while important limitations remain. Gramkow et al. (2025) identified a substantial body of research on wearable and portable digital biomarkers but comparatively limited evidence regarding diagnostic and prognostic applications. Qi et al. (2025) further highlighted insufficient external validation and calibration in AI-based digital biomarker research.
Together, these findings suggest that technological innovation is progressing more rapidly than the accumulation of evidence required for routine clinical implementation.
A central conceptual challenge concerns the distinction between measurement sensitivity and clinical specificity.
Digital sensors can detect subtle variations in behavior or physiology. However, these variations may arise from numerous interacting influences, including neurological conditions, emotional states, environmental demands, medications, and individual behavioral habits.
Greater measurement frequency can improve the characterization of variability, but it does not independently establish the clinical meaning of the observed changes.
Similarly, the ability of machine learning systems to identify statistical patterns should not be confused with the capacity to explain underlying cognitive mechanisms or establish disease-specific diagnoses.
This distinction becomes particularly important in neurodegenerative disease research, where cognitive and behavioral manifestations may overlap across different conditions.
The most plausible near-term application of digital neuropsychology is therefore a complementary model in which digital measures extend established clinical assessment.
Under such a model, conventional neuropsychological examination provides detailed information about cognitive strengths, weaknesses, and diagnostic considerations, while digital monitoring contributes information about temporal variability and functioning in everyday contexts.
AI-based analysis may support the integration of these information sources but should remain subject to clinical oversight and independent validation.
Ultimately, the scientific value of digital neuropsychology will depend less on the sophistication of individual technologies than on their ability to generate reliable, interpretable, equitable, and clinically meaningful information.
Conclusion
Artificial intelligence, smartphone-based cognitive assessment, wearable technologies, and digital biomarkers are contributing to an important transformation in neuropsychological research and clinical practice.
These approaches offer opportunities for repeated cognitive assessment, ecologically relevant measurement, individualized monitoring, and the analysis of complex behavioral and physiological information.
They may prove especially valuable for the early identification of cognitive changes, the longitudinal investigation of neurodegenerative disorders, and the development of personalized assessment and rehabilitation strategies.
However, substantial challenges remain regarding psychometric validation, algorithmic generalizability, data privacy, technological accessibility, clinical interpretability, and regulatory oversight.
Current evidence does not justify the widespread replacement of comprehensive neuropsychological examination with automated digital systems.
Instead, digital neuropsychology should be developed as an evidence-based extension of established clinical methods.
The future of the field will depend on interdisciplinary collaboration among neuropsychologists, neurologists, data scientists, engineers, ethicists, and patients.
Through rigorous validation, transparent artificial intelligence development, and human-centered implementation, digital neuropsychology may contribute to more continuous, personalized, and accessible approaches to understanding cognitive health and neurological disease.
