Curve walking under dual-task conditions combined with machine learning achieved 89.04% accuracy for detecting cognitive impairment vs. healthy controls and 93.02% for distinguishing AD vs. MCI, supporting curved dual-task gait analysis as a promising paradigm for early cognitive impairment detection and staging.
Key Findings
Results
Machine learning models combining single-task and dual-task gait markers achieved high classification accuracy for both cognitive impairment detection and staging.
Models achieved 89.04% accuracy for CI vs. HC classification
Models achieved 93.02% accuracy for AD vs. MCI classification
Models combined both single-task (ST) and dual-task (DT) markers
Results confirmed that curved walking under dual-task conditions provides richer discriminative gait information than single-task curved walking alone
Results
Statistical analysis identified significantly more discriminative gait markers under dual-task conditions than single-task conditions during curve walking.
156 total gait markers were extracted from participants
22 single-task (ST) markers were identified as significantly discriminative
34 dual-task (DT) markers were identified as significantly discriminative
Selected markers spanned macro, micro-temporal, spatial, and symmetry categories
Results
Participants with cognitive impairment demonstrated slower, less stable gait with greater variability and asymmetry, particularly under dual-task conditions.
CI participants showed slower gait compared to healthy controls
Greater gait variability and asymmetry were observed in CI participants
These differences were more pronounced under dual-task (DT) conditions
The dual-task condition involved counting backward from 100 by ones while walking
Methods
The study enrolled 73 older adults across three cognitive groups to perform curve walking under single-task and dual-task conditions captured by a depth camera.
73 older adults were included: healthy controls (HC), MCI due to Alzheimer's disease, and mild AD
Participants walked along a 10-meter oval path
3D body joint motion was captured using a Kinect v.2 depth camera
The dual-task condition consisted of counting backward from 100 by ones while walking
Comprehensive macro- and micro-level gait markers were extracted using signal processing and statistical analysis
Background
Curve walking—especially under dual-task conditions—was hypothesized and confirmed to better expose early CI-related gait changes compared to straight-path walking.
Most prior gait analysis studies focus on straight-path walking, which may miss subtle motor-cognitive deficits
The study specifically used a curved (oval) 10-meter path to increase gait complexity
Dual-task curved walking revealed subtle gait deficits linked to CI that single-task walking did not fully capture
The approach was designed to be low-cost and non-invasive compared to current diagnostic tools such as invasive biomarker testing
Background
The authors identified current cognitive impairment diagnostic tools as limited by invasiveness and cost, motivating gait-based screening approaches.
Early detection of CI, including MCI and Alzheimer's disease, remains limited by the invasiveness and cost of current diagnostic tools
Gait analysis was presented as a promising alternative for non-invasive screening
The depth camera (Kinect v.2) system was described as enabling low-cost, non-invasive data capture
The authors note the approach supports potential use as a screening tool for supervised clinical and community settings, subject to validation in less-controlled environments
What This Means
This research suggests that having older adults walk along a curved path while simultaneously performing a mental task (counting backward) can reveal subtle walking differences that distinguish people with healthy cognition from those with mild cognitive impairment (MCI) or Alzheimer's disease. The study used a low-cost depth camera (similar to gaming motion-capture technology) to measure detailed 3D movements of 73 older adults as they walked an oval path. By extracting 156 different measurements of how people walk—including step timing, body symmetry, and spatial patterns—the researchers found that the combination of curved path and mental challenge uncovered more meaningful differences between groups than walking alone did, identifying 34 useful markers under dual-task conditions compared to only 22 under single-task conditions.
When these walking measurements were fed into machine learning computer programs, the system could correctly identify people with cognitive impairment versus healthy controls about 89% of the time, and could distinguish between MCI and Alzheimer's disease about 93% of the time. People with cognitive impairment walked more slowly, less steadily, and with greater unevenness between their left and right sides, and these differences became especially clear when they were also doing the mental counting task. This suggests that adding a cognitive challenge to a structured walking test makes it a more powerful tool for detecting brain-related problems.
This research is significant because current methods for diagnosing Alzheimer's disease and related conditions often involve expensive brain scans, spinal taps, or lengthy neuropsychological testing. This research suggests that a relatively inexpensive depth camera combined with a specific walking protocol and computer analysis could serve as an accessible screening tool in clinics or community settings. The authors note that further testing in real-world, less-controlled environments would be needed before this approach could be broadly adopted.
Check Your Own Numbers
Upload your bloodwork. We'll cross-reference your results against this study and 4,700 others.
Seifallahi M, Galvin J, Ghoraani B. (2026). Curve walking gait analysis using depth camera and machine learning for detecting cognitive impairment.. Frontiers in neurology. https://doi.org/10.3389/fneur.2026.1777635