Dementia Brain Rewires Itself to Keep Stepping Fast, fNIRS Study Reveals

Every year, falls send millions of older adults to the hospital, and for those living with dementia the danger is dramatically amplified. People with dementia face roughly twice the fall risk of their cognitively healthy peers and suffer three times the rate of serious injury, and more than half of these falls begin with an unexpected slip or trip that demands a split-second decision to move the right foot to the right place. A new study published in GeroScience has now peered inside the living brain to ask what happens, at the level of neural circuitry, when the aging brain tries to make that decision quickly. The answer reveals a surprising form of neural improvisation: rather than simply breaking down, the prefrontal cortex of people with mild-to-moderate dementia appears to reorganize itself into more tightly clustered local networks, and the degree of that reorganization tracks with how fast they can step.

The research, led by Jae Q. J. Liu and Wayne L. S. Chan of The Hong Kong Polytechnic University together with colleagues at The Education University of Hong Kong and Neuroscience Research Australia, focused on a deceptively simple behavioral measure called choice stepping reaction time, or CSRT. In the clinical test, a participant stands on a nonslip mat with four stepping panels arranged to the left, right, left-up, and right-up. On hearing a verbal cue, they must step onto the indicated panel as quickly and accurately as possible and return to center. The total time to complete twelve correct steps is recorded. CSRT is not merely a test of leg strength; it is a composite measure of sensory processing, attention, decision-making, and motor execution, and it has been established as an independent predictor of recurrent falls in older adults. Previous work has shown that individuals with mild-to-moderate dementia take nearly twice as long as healthy older adults to complete the task, yet the neural machinery underlying this slowdown had remained largely unexplored.

To capture that machinery in action, the team recruited twenty-six healthy older adults and twenty-five older adults with mild-to-moderate dementia, all aged sixty-five or above, all right-handed, and all able to walk ten meters independently. Dementia participants carried a clinical diagnosis under DSM-5 criteria and scored between five and eighteen on the Hong Kong version of the Montreal Cognitive Assessment, while healthy controls scored twenty-two or above; anyone falling in the borderline nineteen-to-twenty-one range was excluded to avoid mild cognitive impairment contaminating either group. Participants wore a portable, high-density functional near-infrared spectroscopy system, a 48-channel device known as NIRSIT that shines near-infrared light at 780 and 850 nanometers through the forehead to track oxygenated hemoglobin, the blood-oxygen signal that rises when a brain region works harder. Because fNIRS is wearable and tolerant of movement, it allowed the researchers to measure brain activity while people actually stood and stepped, something conventional MRI scanners cannot easily accommodate.

The experimental design contrasted two stepping conditions. In the simple stepping task, participants repeatedly stepped to the same target panel, so response selection was required only at the start of each block. In the choice stepping task, auditory cues were randomized trial by trial, forcing continuous response selection and sustained attention across all eight trials in each of four blocks per condition. Blocks lasted forty seconds, separated by randomized rest periods of quiet standing, and the entire protocol took roughly eight minutes. The researchers then examined oxygenated hemoglobin responses across eight prefrontal subregions and, crucially, went a step beyond traditional activation analysis by constructing functional connectivity matrices among prefrontal channels and applying graph theory, the mathematical language of networks, to quantify how the prefrontal cortex was wired together during each task.

The activation results told a striking story of constrained neural resources. Healthy older adults showed the expected pattern: right dorsolateral prefrontal cortex activity increased when the task shifted from simple to choice stepping, consistent with the compensation-related utilization of neural circuits hypothesis, which holds that the aging brain recruits extra neural resources as cognitive demands rise. The dementia group did the opposite. They showed greater right dorsolateral activation than controls during the simple task, yet their activation actually decreased when demands rose to the choice condition. In the task-onset window, the divergence was even sharper: healthy participants ramped up right dorsolateral activity at the start of choice stepping, while the dementia group showed no such modulation. The authors interpret this through the CRUNCH framework, which predicts that a brain with diminished neural reserve hits the peak of its activation-load curve prematurely and then declines under further load. For these participants, even simple stepping may already have pushed the prefrontal cortex near its ceiling.

But the most provocative findings emerged from the network analysis. Using the GRETNA toolbox, the team computed four global graph metrics across a sparsity range of 0.10 to 0.50: global efficiency, local efficiency, characteristic path length, and the clustering coefficient, which measures the tendency of nodes to form densely interconnected local clusters. Only the clustering coefficient distinguished the groups. Older adults with dementia showed significantly higher prefrontal clustering during both stepping tasks, indicating that their prefrontal networks had become more segregated, organized into tighter local modules with less distributed integration. Global efficiency, local efficiency, and path length were statistically comparable between groups, meaning the dementia brain had not simply collapsed into random disorganization. Instead, within the prefrontal cortex specifically, the architecture had shifted toward local specialization.

Even more remarkable was what that segregation predicted behaviorally. Within the dementia group, higher prefrontal clustering during choice stepping correlated with faster choice stepping reaction time, with a correlation coefficient of negative 0.552 that survived false discovery rate correction. A multiple linear regression adjusting for age, education, and MoCA score showed that a 0.01-unit increase in clustering coefficient was associated with a 14.32-second decrease in the twelve-step completion time, and adding clustering to a covariate model improved the explained variance in CSRT by 35.6 percentage points. No such relationship existed in the healthy group, and none of the other network metrics predicted stepping speed in either cohort. The authors suggest this reflects a compensatory scaffold, consistent with the revised scaffolding theory of aging and cognition: as neurodegeneration erodes global network integration and amplifies neural noise, the prefrontal cortex may partially offset the damage by tightening local subcircuits, preserving enough executive function to support rapid volitional stepping.

The study also uncovered an intriguing signature of healthy aging. Among the cognitively intact participants, advancing age was associated with greater global efficiency, greater local efficiency, and shorter characteristic path length during stepping, a pattern the researchers read as compensatory architectural refinement that keeps motor-cognitive control intact. Notably, this age-efficiency coupling vanished entirely in the dementia group, suggesting that dementia pathology derails the natural aging trajectory of prefrontal network function. An exploratory analysis of interregional functional connectivity found no significant group or task effects after correction, reinforcing that the key dementia-related signal lives at the level of whole-network topology rather than in any single prefrontal connection.

The clinical implications could be substantial. The auditory-cued stepping paradigm is cognitively accessible and relatively independent of education level, making it practical for dementia populations, and pairing it with fNIRS could yield neural biomarkers for stratifying fall risk before a fall ever happens. The findings also raise a testable hypothesis about intervention: step-based training programs have already demonstrated a 26 percent reduction in fall incidence in randomized trials, and the new results suggest those benefits might operate partly through recruiting a prefrontal compensatory scaffold. Precision neuromodulation techniques such as high-definition transcranial direct current stimulation and intermittent theta-burst stimulation, which have been shown to modulate network segregation, could conceivably be tuned to strengthen this adaptive reorganization. The authors caution that their sample was small, predominantly female, and age-mismatched between groups, and that the cross-sectional design cannot establish causation. Longitudinal studies will be needed to determine whether prefrontal topology is modifiable and whether modifying it delays the slide toward network randomness. Still, the core message stands: in the dementia brain, the wiring itself may be fighting back, and measuring that fight could change how we protect the mobility of millions.

Subject of Research: Prefrontal network topology reconfiguration during volitional stepping in aging and dementia

Article Title: The choice stepping reaction task: how prefrontal network topology reconfigures in aging and dementia

Article References: Liu, J. Q. J., Yeung, M. K., Chen, M., Menant, J. C., Lord, S. R., Sturnieks, D. L., Tam, Y. K., Tang, P. M., Wong, K. C., Wong, K. L., Wong, Y. L., & Chan, W. L. S. (2026). The choice stepping reaction task: how prefrontal network topology reconfigures in aging and dementia. GeroScience. https://doi.org/10.1007/s11357-026-02534-y

Image Credits: AI Generated

DOI: 10.1007/s11357-026-02534-y

Keywords: dementia, prefrontal cortex, fNIRS, choice stepping reaction time, graph theory, network topology, fall risk, cognitive aging, compensatory scaffolding, GeroScience, motor control, neurodegeneration