Skip to main content
Talk

Deep Learning and Multimodal Sensing for Early Dementia Research — NeuroDPlatform

Dr. Omar Costilla Reyes presented his doctoral research on deep learning and multimodal sensing for dementia at NeuroDPlatform on March 8, 2017. The session took place at the University of Manchester.

Doctoral research on multimodal sensor data and early dementia detection presented at a NeuroDPlatform session, University of Manchester, March 8, 2017. Credit: Image courtesy of Dr. Omar Costilla Reyes.

A doctoral project presented to a multidisciplinary platform

On March 8, 2017, Dr. Omar Costilla Reyes presented his PhD project at a NeuroDPlatform session at the University of Manchester, under the title “Deep Learning and Multimodal Sensing for Early Dementia Research.” He was then a third-year doctoral candidate in the School of Electrical and Electronic Engineering, and he described the session as a chance to contribute to the multidisciplinary synergies the platform was created to enable. A NeuroD banner visible in the photograph on this page asks researchers whether tools, techniques, or algorithms from their own fields could be applied to neurodegenerative diseases.

Sensing, deep learning, and early dementia

The slide shown in the photograph is titled “Analysis of multimodal human-generated sensor data: a data science approach for early detection of dementia.” Human-generated sensor data are the measurements that arise as people go about ordinary activities, such as the footsteps recorded as someone walks across a floor sensor. Multimodal means that more than one kind of measurement is combined, which offers a broader picture of behavior than any single signal. Deep neural networks are learning methods that can find useful features in raw signals without hand-designed descriptors, so they suit this kind of data. Dementia often begins with subtle changes in how people move and think long before it is recognized in the clinic, so the project asks how patterns in everyday sensor recordings can help researchers understand and detect those early changes.

Where the work led

The project is documented in his University of Manchester thesis, which includes studies of how cognitively demanding tasks affect gait patterns recorded by a floor sensor. Related results were presented as a poster at the 2017 Advances in Data Science Conference and at ESOF 2018 in Toulouse, and published in “Age-Related Differences in Healthy Adults Walking Patterns Under a Cognitive Task With Deep Neural Networks” (2020). Together they show that walking recorded by a floor sensor carries information about how the brain is engaged, and that deep neural networks can read it, which gives dementia research a new kind of measurement to build on.

Topics