AI-Powered Dementia Detection: The Promise of Early Intervention
From PENSIEVE-AI's 0.93 AUC in five-minute drawing-based screenings to novel biomarker detection, artificial intelligence is transforming the race to detect and intervene in dementia before irreversible damage occurs.
The Dementia Crisis: Scale and Urgency
Dementia represents one of the most devastating and costly health challenges of the twenty-first century. The World Health Organization estimates that more than 55 million people worldwide currently live with dementia, a number projected to reach 78 million by 2030 and a staggering 139 million by 2050 as global populations age. Alzheimer's disease, the most common form of dementia, accounts for 60-70% of cases and currently has no cure. The global economic burden of dementia exceeds $1.3 trillion annually, encompassing direct medical costs, social care costs, and the enormous unpaid labor of family caregivers who provide the majority of day-to-day support.
The cruelest aspect of dementia is its insidious onset. By the time most patients receive a diagnosis, the underlying neurodegenerative process has been progressing for 15-20 years. Neurons have already been lost, synaptic connections have deteriorated, and the window for the most effective interventions has significantly narrowed. Current diagnostic methods, which rely on neuropsychological testing, expensive neuroimaging, and invasive cerebrospinal fluid analysis, are typically deployed only after symptoms become obvious enough to prompt a clinical evaluation, missing the critical early stages when emerging therapies show the greatest promise.
This diagnostic gap represents both a tragedy and an opportunity. If AI-powered screening tools can detect the earliest signs of cognitive decline during routine interactions, years before symptoms become clinically apparent, they could fundamentally alter the trajectory of dementia care by enabling intervention during the window when treatments are most likely to slow or halt disease progression.
PENSIEVE-AI: Five-Minute Drawing Test with 0.93 AUC
Among the most promising developments in AI-powered dementia detection is PENSIEVE-AI, developed in Singapore through a collaboration between Singapore General Hospital (SGH) and the Government Technology Agency of Singapore (GovTech). Rather than relying on speech or interviews, PENSIEVE-AI uses a brief digital drawing test: participants complete four drawing tasks, including shapes and a clock, on a touch-screen tablet with a stylus. In a prospective validation study of 1,758 community-dwelling adults aged 65 and older, the deep-learning model detected clinically adjudicated mild cognitive impairment and dementia with an area under the curve (AUC) of 0.93, comparable to traditional, far longer neuropsychological assessments (AUC 0.94).
The technical approach underlying PENSIEVE-AI analyzes how a person draws, not what they say. Deep-learning models extract features from the drawing process, capturing spatial accuracy, motor execution, and the way each figure is constructed on the tablet. Because performance depends on drawing rather than reading or verbal fluency, the test is largely independent of a participant's education and literacy, an important property in linguistically diverse populations. The result is a screening tool that is not only accurate but accessible, requiring only a tablet and a few minutes of the user's time; most participants completed it in under five minutes, with a median completion time of 3.7 minutes.
The accessibility of PENSIEVE-AI's approach is as significant as its accuracy. Traditional dementia screening requires a visit to a specialist clinic, administration of standardized cognitive tests by trained professionals, and often expensive imaging studies. These barriers mean that screening is typically limited to patients who have already shown obvious symptoms or who have the resources and motivation to seek evaluation proactively. A largely self-administered, tablet-based test that can be deployed in primary care or community settings removes these barriers and makes population-scale early detection feasible; the developers have indicated the tool is slated for wider rollout in Singapore from 2026.
Beyond Cognitive Testing: Novel Biomarker Detection
While drawing and cognitive analysis represent the most mature approaches to AI-powered dementia detection, researchers are exploring a broader range of biomarkers that AI systems can detect non-invasively. In one line of work, researchers at the Chinese University of Hong Kong, the National University of Singapore, and the Singapore National Eye Centre used a deep-learning system to measure retinal blood-vessel calibre from standard retinal photographs. In a prospective cohort followed for up to five years, narrower retinal arteriolar calibre at baseline was associated with an increased risk of cognitive decline. The retinal vasculature shares developmental origins with cerebral blood vessels, making the eye a non-invasive window into brain vascular health, and AI analysis of routine retinal photographs can surface microvascular changes that correlate with neurodegenerative risk.
Gait analysis represents another promising biomarker domain. Changes in walking patterns, including reduced speed, increased variability, shorter stride length, and altered balance, have been shown to precede cognitive decline by several years. AI systems that analyze gait through video, wearable sensors, or ambient floor sensors can continuously monitor these subtle changes without requiring active participation from the individual being monitored. In senior living communities where ambient monitoring is already deployed for fall prevention, gait analysis for cognitive health screening represents a natural extension of existing infrastructure.
Sleep pattern analysis is emerging as yet another domain where AI-detected changes can signal early neurodegeneration. Disrupted sleep architecture, including reduced slow-wave sleep, increased sleep fragmentation, and altered REM sleep patterns, is now understood to be both a consequence and a contributor to neurodegenerative processes. AI analysis of data from consumer sleep tracking devices, smart mattresses, or ambient bedroom sensors can detect sleep pattern changes that may indicate early cognitive decline, adding another layer to the multi-modal early detection approach that researchers increasingly view as the most promising path forward.
Intervention at the Earliest Stages
Early detection is valuable only if it enables meaningful intervention. The landscape of dementia therapeutics is evolving rapidly, with several anti-amyloid antibody therapies receiving approval or approaching approval for early-stage Alzheimer's disease. Lecanemab, approved by the FDA in 2023, demonstrated a 27% slowing of cognitive decline in patients with early Alzheimer's disease. Donanemab, approved in 2024, showed a 35% slowing in a subset of patients treated early in their disease course. These therapies are most effective when initiated during the earliest detectable stages of disease, making early AI-powered screening directly relevant to treatment outcomes.
Beyond pharmaceutical interventions, a growing body of evidence supports the effectiveness of lifestyle modifications in reducing dementia risk and slowing early-stage progression. The FINGER trial and subsequent multi-domain intervention studies have demonstrated that combined programs of physical exercise, cognitive stimulation, dietary modification, and cardiovascular risk management can significantly reduce cognitive decline in at-risk populations. AI-powered health monitoring systems can support these lifestyle interventions by tracking adherence, adapting recommendations based on individual progress, and providing the kind of sustained, personalized engagement that makes long-term behavior change achievable.
The combination of early AI detection and comprehensive intervention support creates a new paradigm for dementia care: instead of waiting for symptoms to become severe enough for clinical diagnosis, proactively screening populations, identifying at-risk individuals during the preclinical window, and delivering personalized intervention programs that combine pharmaceutical and lifestyle approaches. This paradigm shift has the potential to significantly reduce the incidence and severity of dementia across aging populations.
Scaling AI-Powered Dementia Screening Globally
Scaling AI-powered dementia detection from clinical research to population-level deployment requires addressing several challenges. Cultural and linguistic diversity demands that screening tools be validated across different languages, dialects, and cultural contexts. Educational background, literacy levels, and familiarity with technology all influence cognitive test performance and must be accounted for to avoid systematic bias. PENSIEVE-AI's validation across Singapore's diverse, multi-literacy population provides a model for this kind of inclusive validation, but extending this approach to the full diversity of global aging populations remains an ongoing challenge.
Data privacy and consent present particular challenges in dementia screening. Individuals being screened for cognitive decline may have diminished capacity to provide informed consent for data collection and analysis. Screening results carry significant personal and social implications, affecting everything from driving privileges to financial decision-making authority. AI-powered screening systems must be designed with robust privacy protections, clear consent processes, and careful governance of how results are communicated and used.
Ajentik is contributing to the global effort to scale AI-powered dementia screening through our health monitoring platform's cognitive assessment capabilities. Our voice AI agents can conduct conversational cognitive screenings during routine interactions, analyzing the linguistic and cognitive features that correlate with early decline without requiring dedicated assessment sessions. By embedding screening into the natural flow of daily AI-assisted interactions, we aim to make early detection as routine and unobtrusive as checking blood pressure. Combined with our care coordination agents that can connect flagged individuals with appropriate clinical follow-up, our platform provides an end-to-end pathway from early detection to timely intervention.
Sources
- World Health Organization, "Dementia Fact Sheet," 2024
- Liew TM et al., "PENSIEVE-AI a brief cognitive test to detect cognitive impairment across diverse literacy," Nature Communications, 23 March 2025, https://www.nature.com/articles/s41467-025-58201-x
- Cheung CY et al., "Deep-learning retinal vessel calibre measurements and risk of cognitive decline and dementia," Brain Communications, vol. 4(4), 17 August 2022, https://academic.oup.com/braincomms/article/4/4/fcac212/6669988
- FDA, "Lecanemab Approval for Early Alzheimer's Disease," 2023
- The Lancet Commission on Dementia Prevention, Intervention, and Care, 2024 Update
- FINGER Trial Consortium, "Multi-Domain Intervention Results: Five-Year Follow-Up," 2024
Related Articles
How Agentic AI Is Revolutionizing Elderly Care in 2026
Autonomous AI agents are transforming senior care through intelligent monitoring, meaningful companionship, and seamless care coordination.
Source: World Health Organization, "Ageing and Health" Fact Sheet, 2024
Healthcare AIAI Companions for Seniors: Promise, Evidence, and the Case for Caution
MIT Technology Review named AI companions a 2026 breakthrough while warning of real harms. For older adults, purpose-built companion robots are showing measurable, government-documented benefits, if deployed responsibly.
Source: MIT Technology Review, "AI companions: 10 Breakthrough Technologies 2026," January 12, 2026 — https://www.technologyreview.com/2026/01/12/1130018/ai-companions-chatbots-relationships-2026-breakthrough-technology/
TechnologyVoice-First AI: Unlocking Senior Engagement Through Natural Conversation
As voice-based AI interfaces mature, they are proving to be the most natural and accessible modality for engaging older adults with technology that genuinely improves their daily lives.
Source: AARP, "2025 Tech Trends: Tech Adoption Continues Among Older Adults," 2025 — https://www.aarp.org/pri/topics/technology/internet-media-devices/2025-technology-trends-older-adults/
Building with Agentic AI?
Learn how Ajentik's autonomous agent platform is helping enterprises deploy production-ready AI agents at scale.
Schedule a Consultation