Abstract
Path integration (PI) is impaired early in Alzheimer's disease (AD) but reflects multiple sub-processes that may be differentially sensitive to AD. To characterize these sub-processes, we developed a novel generative linear-angular model of PI (GLAMPI) to fit the inbound paths of healthy elderly participants performing triangle completion, a popular PI task, in immersive virtual reality with real movement. The model fits seven parameters reflecting the encoding, calculation, and production errors associated with inaccuracies in PI. We compared these parameters across younger and older participants and patients with mild cognitive impairment (MCI), including those with (MCI+) and without (MCI-) cerebrospinal fluid biomarkers of AD neuropathology. MCI patients showed overestimation of the angular turn in the outbound path and more variable inbound distances and directions compared with healthy elderly. MCI+ were best distinguished from MCI- patients by overestimation of outbound turns and more variable inbound directions. Our results suggest that overestimation of turning underlies the PI errors seen in patients with early AD, indicating specific neural pathways and diagnostic behaviors for further research.