Overview: A New Path in Dementia Screening
As the global population ages, dementia research is shifting toward early, noninvasive biomarkers that can be easily integrated into clinical workflows. The Sensorized Motor and Cognitive Dual Task Framework (MCDT) for dementia diagnosis presents a promising approach by combining motor assessments with cognitive load, all captured via wearable sensors. This cross-sectional study investigates whether upper-limb motor tasks (ULMF) and lower-limb motor tasks (LLMF) can outperform traditional walking dual tasks (GAIT) in differentiating older adults with normal cognition, subjective cognitive impairment (SCI), and mild cognitive impairment (MCI).
What is the MCDT Framework?
The MCDT paradigm leverages motor-cognitive interference: performing a movement while engaging in a cognitive task taxes neural networks, revealing subtle cognitive decline that may not be evident in standard tests. In this study, five motor-cognitive tasks were examined: two upper-limb tasks (forefinger tapping FTAP and thumb-forefinger tapping THFF), two lower-limb tasks (toe tapping TTHP and heel tapping HTTP), and the gold standard walking task (GAIT). Each task was performed as a single task (ST) and under three cognitive load levels (CL0 to CL3), with CL1, CL2, and CL3 representing progressively challenging counts backward by 1, 3, and 7 respectively.
Methods at a Glance: Wearable Sensors and Data Fusion
The study deployed a low-cost, noninvasive MEMS-based wearable system consisting of four IMU devices (two for hands and two for feet). Data were streamed wirelessly and analyzed offline to extract 45 features across the five MCDTs. The researchers computed a weighted dual-task cost (DTC) that accounts for cognitive performance by incorporating the number of correct responses, thus capturing cognitive efficiency and commitment. To counter the study’s small sample size, they created five pooled indices by aggregating related DTC components, which then fed into logistic regression models to classify participants into OA, SCI, or MCI categories. Age and FAB (Frontal Assessment Battery) scores were included to control for age effects and executive function, respectively.
Key Findings: ULMF vs LLMF Performance
The results reveal nuanced differences between upper- and lower-limb tasks in multi-class classification. In the two-class problem (MCI vs OA), several MCDT models based on THFF (ULMF) and the lower-limb tasks (LLMF) outperformed the GAIT benchmark. Specifically, THFF and HTTP achieved the highest accuracy (up to 93%), indicating that increased motor complexity and dual-task demands enhance discriminative power for MCI detection. In contrast, FTAP hovered around GAIT performance, suggesting that single-finger tasks may be less sensitive to cognitive decline when distinguishing MCI from healthy aging.
When expanding to a three-class problem (OA vs SCI vs MCI), lower-limb tasks (TTHP and HTTP) continued to show strong performance, with TTHP often outperforming HTTP. Hand-based tasks provided valuable information for identifying OA, potentially due to the greater motor planning demands of fine hand movements. Overall, the findings suggest a broader MCDT repertoire—encompassing both ULMF and LLMF tasks—offers improved diagnostic insight, especially for distinguishing MCI from SCI and OA.
Clinical Implications and Next Steps
The study supports the feasibility of a single, IMU-based wearable system to conduct a spectrum of MCDTs in clinical settings, research labs, and potentially home environments. This framework could enable rapid, noninvasive screening and monitoring of cognitive decline, facilitating earlier interventions. Importantly, the approach emphasizes adaptive task selection: higher cognitive load and motor complexity may yield greater diagnostic separation, particularly for MCI and SCI. Future work should validate these results in larger, more diverse cohorts and explore longitudinal trajectories to assess progression from SCI to MCI and beyond. Integrating a broader neuropsychological battery would also enrich understanding of how executive function, memory, language, and visuospatial skills interact with motor-cognitive interference.
Limitations and Considerations
As a cross-sectional study with 44 participants, generalizability is limited. The authors call for expanding OA sample sizes, stratifying MCI subtypes, and refining motor-load levels to delineate subtle differences among groups. Additional longitudinal data will help determine whether MCDT-based indices can predict conversion to dementia.
Conclusion: A Promising, Accessible Biomarker Platform
The Sensorized MCDT framework demonstrates that integrating upper- and lower-limb motor tasks with cognitive challenges—captured through wearable IMUs—can yield accurate, noninvasive differentiation among OA, SCI, and MCI. By embracing a diversified task set and robust analytics, this approach could become a practical tool for early dementia screening and progression monitoring.
