Poster 51 at the EuroScience Open Forum
The EuroScience Open Forum (ESOF) is a biennial, interdisciplinary meeting on science and innovation, and its 2018 edition, the eighth, took place in Toulouse, France, in July. Dr. Omar Costilla Reyes, of the University of Manchester, presented “Early detection of dementia with ambient sensing and deep machine learning” as poster 51 in the student competition. He announced the poster in June, with thanks to the university’s Data Science Institute for its support, and invited visitors to the poster area during the forum.
The poster framed the work around a gap in knowledge: the link between patterns of human gait and cognitive-executive decline is not well understood. Ambient sensing means measuring people through sensors built into the environment, here a floor sensor, rather than through devices they wear. The study examined dual-task walking in a group of healthy adults, using raw floor-sensor measurements rather than subjective gait features such as velocity or variability.
What the poster reported
In a dual-task experiment a person walks while performing a second, cognitively demanding activity, and researchers look at how the walking pattern changes. The poster describes a spatio-temporal deep-learning method that learns from both the position and the timing of footsteps in the floor-sensor data. It used t-SNE, a technique that projects high-dimensional data onto a two-dimensional map, to visualize how the trained network separated walking patterns. The poster reports a very high F-score, a combined measure of precision and recall, for identifying dual-task walking, well above that of a classical machine-learning approach in the best case.
The poster presents these results as a research approach for studying the effects of mild cognitive impairment and for finding potential biomarkers of neurodegenerative conditions, including prodromal Alzheimer’s disease, the stage before dementia is diagnosed. Because the participants were healthy adults, the study shows how the method can pick up subtle differences in how people walk under cognitive load, which is exactly the kind of signal that research on early cognitive change looks for.
Connections to related work
The poster belongs to the Manchester floor-sensor research line, supported by a Data Science Institute small project on gait patterns and early cognitive decline and later published with the same collaborators as “Age-Related Differences in Healthy Adults Walking Patterns Under a Cognitive Task With Deep Neural Networks.” At the same forum he also spoke with the mathematician Cédric Villani, an exchange described on its own page, and he had presented a different gait poster at ESOF 2016 in Manchester.