Past Event! Note: this event has already taken place.
Machine Learning for Medical Data
February 16, 2018 at 2:00 PM to 3:30 PM
|Location:||5345 Herzberg Laboratories|
With the advances in low-cost wearable devices and miniaturized sensors, wide spread internet access on smartphones and tablets, it is increasingly easy to gather large amounts of health data. This opens up new possibilities in different applications ranging from monitoring and diagnosing diseases to security and access control.
In the first part, I will describe the opportunities and challenges associated with the application of physiological signals such as electrocardiogram (ECG) for human recognition as well as our patent pending approach for biometric spoof detection. Dealing with such signals is challenging because they are time varying and their waveform are affected by factors such as emotion, diet and heartrate, specially when recorded in different recording sessions. I will describe the localized feature selection method that I have developed for such non-stationary data. It uses local information in terms of sample margins while enforcing an across-session measure to cope with non-stationarity of physiological signals. Also, I will briefly describe the novel method that I have developed for cross-view matching on multiview data.
In the second part, I will describe how machine learning and natural language processing can help in assessment of neurodegenerative disorders such as Alzheimer’s disease (AD). There is increasing evidence that linguistic aspects of speech, such as the ratio of pronouns to nouns, relate strongly to cognitive decline. Indeed, language is one of the earliest capacities afflicted by AD. We have extracted hundreds of linguistic measures including lexico-syntactic, acoustic, and semantic features form short speech samples for assessment of AD. Speech samples are recorded via an automated web-/telephony- based system that remotely performs some language tests similar to those found in standard tests for cognitive ability and language impairment such as picture description or story retelling.
Majid Komeili is a postdoctoral fellow at University of Toronto working jointly in Department of Computer Science, Toronto Rehabilitation Institute and Vector Institute. He received a Ph.D. from Department of Electrical and Computer Engineering at University of Toronto in September 2017. He received a M.Sc. from Tarbiat Modares University, Iran, in 2008, and a B.Sc. from Iran University of Science and Technology in 2006. His research interests include machine learning, natural language processing, physiological signal analysis, medical biometrics and data analytics for healthcare.