Skip to main content
Talk

Analysis of Multimodal Human-Generated Sensor Data: A Deep Learning Approach for Early Detection of Dementia

In May 2017, Dr. Omar Costilla Reyes presented a poster on multimodal sensor data and deep learning for early dementia detection. It was shown at the University of Manchester’s Advances in Data Science Conference.

Poster on multimodal sensor data and deep learning for early dementia detection at the Advances in Data Science Conference, University of Manchester, May 2017. Credit: Image courtesy of Dr. Omar Costilla Reyes.
Poster-session program for the University of Manchester’s Advances in Data Science Conference 2017, listing the multimodal sensor-data poster first. Credit: University of Manchester Data Science Institute; image courtesy of Dr. Omar Costilla Reyes.

Poster at a two-day data-science meeting

The University of Manchester’s Data Science Institute organized the Advances in Data Science Conference in May 2017 as a two-day meeting on recent developments in machine learning, Bayesian statistics, scalable algorithms, privacy, visualization, and applications, with speakers from industry and academia. Dr. Omar Costilla Reyes, then a doctoral researcher in the School of Electrical and Electronic Engineering, presented the poster “Analysis of multimodal human-generated sensor data: A deep learning approach for early detection of dementia,” listed first in the conference’s poster program with Dr. Patricia Scully and Prof. Krikor Ozanyan as co-authors. He announced beforehand that he would take part and invited attendees at the conference to find the work in the poster area. The conference was organized by Magnus Rattray and Neil Lawrence.

What the poster set out

The poster is laid out as a path from problem to results. It starts from the difficulty that Alzheimer’s disease, a progressive age-associated neurodegenerative disorder, is often diagnosed only at a moderate or advanced stage, and that the biomarkers available are often expensive or invasive. It proposes instead to look for inexpensive, non-invasive “disease signatures” in data that people produce unobtrusively in their own homes, such as patterns of computer use and the way they walk. For the walking part it used a floor sensor and a convolutional neural network, a deep-learning model that learns features directly from raw sensor signals, and it reported very high accuracy in classifying walking activities on the floor-sensor data. The poster presented this result as research progress toward early detection of dementia from everyday sensor data.

Related research

The poster connected to his floor-sensor studies of walking under cognitive load, later published as “Age-sensitive differences in single and dual walking tasks from footprint floor sensor data” (2017) and “Age-Related Differences in Healthy Adults Walking Patterns Under a Cognitive Task With Deep Neural Networks” (2020), and to a Data Science Institute small-project grant on gait patterns for early detection of cognitive decline awarded in January 2018. The same research line was presented at NeuroDPlatform earlier in 2017.

Topics