UT BIOME
  • Home
  • ABOUT US
  • PROJECTS
  • NEWS & EVENTS
  • RESOURCES
  • CONTACT US
  • Home
  • ABOUT US
  • PROJECTS
  • NEWS & EVENTS
  • RESOURCES
  • CONTACT US

COMPUTATION TEAM
AI Model to Predict Senior Falls Using Wearable Sensor

Project Members: Dhana Abdo, Gerry Chen, Anne Jing, Katherine Zhu (formerly)

Summary

For Canadians, falls are a leading cause of injury-related hospitalizations for those aged 65 or older. Upon completion of this project, we hope to have independently built a working prototype with a fall detection confidence level of 90%. Our goal is to create a working solution to the problem presented that has potential to grow. We hope to grow both individually and as team members.

The Problem

For Canadian seniors, falls are a leading cause (85%) of injury-related hospitalizations for those aged 65 or older [1][2]. Statistics indicate that 20 to 30 percent of seniors fall each year, where each fall harms not only the injured individual, but affect family, friends, care providers, and the health care system [1]. The Government of Canada emphasizes the increasing need for effective falls prevention initiatives [1], as falls not only lead to negative physical consequences (broken/fractured bones), but also mental health outcomes like fear of falling, loss of autonomy and greater isolation (increasing the risk of depression) costing the government an estimated $2 billion annually [1].

Many factors contribute to a fall. As age increase, an individual's coordination decline along with adequate concentration and cognitive functions required to prevent falls and maintain gait [2]. 
Decreases in gait velocity and step length, a wider gait base, and decreases in lower limb strength occur with aging [2] all increase the risk of falling when walking on irregular surfaces. Furthermore, accumulating medical problems along with side-effects from associated medications increase the risk [2]. 

Current solutions do provide an alert to loved ones after a fall, either through a button or automatic detection, though a fall would have already occurred to inflict physical and mental injuries to the victims.
Full physical assessment solutions include gait symptom detection, movement disorder application, and fall detection focus solutions. These are only conducted when one's risk of falling reached 50%. Current gait analysis 
include one from Kinesis Health Technologies with their Kinesis Gait technology, which uses wireless inertial sensors to calculate numbers and generate a mobility assessment [3]. For movement disorder application: smart insoles for shoes consists of various sensors spread about to detect changes in pressure as one walks [4]. Lifeline is a device by Philips with AutoAlert features that constantly measures changes in height orientation to a horizontal position and velocity of an individual [5].

Our Solution

As full physical assessments are only conducted when one's risk of falling reaches 50%, those at lower risks are not continuously monitored and prevention methods tend to be incorporated after a fall. The UT BIOME Computation Team would like to begin with a fall-detection device, but aims to provide a fall prevention device in the long term. The design focus is a simpler model, addressed to falling initially. This allows the Computation Team to work more on developing an accurate model: the algorithm must achieve a confidence level greater than 90%. Measured qualitatively, another main focus of this project is to strengthen the team's knowledge in Arduino Uno, Machine Learning, and its biomedical applications. This helps to place a larger emphasis on the deep learning aspect of this project, while allowing opportunities to extend the design later on. 

​The project will be split into five stages: 

Stage 1 - Designing the Circuitry for Data Collection aims to have the circuit design to be completed and all components should be soldered on to the Arduino Board. 

Stage 2 - Collecting Initial Data will have team members use the developed circuit deign to collect data through performing daily activities. 

Stage 3 - Creating the Classification Model will be completed by referencing existing models and building upon them using additional collected data. 

Stage 4 - Tuning the Hyperparameters of the model to Improve Accuracy aims to adjust the initial classification model to satisfy the Accuracy objective.

 Stage 5 - Add-On Stage, will occur if time permits and will focus on incorporating other features that further build upon the main objective of the design.

Training

  • IEEE's Arduino Uno Workshop
  • UT BIOME's Machine Learning Workshop
  • Tri-Council Policy Statement: Ethical Conduct for Research Involving Humans: Course on Research Ethics
references
[1] P. H. A. of Canada, “Government of Canada,” Canada.ca, 10-Apr-2014. [Online]. Available: https://www.canada.ca/en/public-health/services/health-promotion/aging-seniors/publications/pu blications-general-public/seniors-falls-canada-second-report.html. [Accessed: 17-Sep-2020].
[2] T. Al-Aama, “Falls in the elderly: spectrum and prevention,” Canadian family physician Medecin de famille canadien, Jul-2011. [Online]. Available: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3135440/. [Accessed: 17-Sep-2020].

[3] B. Greene, “Gait assessment,” Kinesis Health Technologies. [Online]. Available: https://www.kinesis.ie/gait/. [Accessed: 17-Sep-2020].
[4] Wang, Changwon et al. “Soft-Material-Based Smart Insoles for a Gait Monitoring System.” Materials (Basel, Switzerland) vol. 11,12 2435. 30 Nov. 2018, doi:10.3390/ma11122435
[5] “Medical Alert Systems for Seniors With Fall Detection: Fall Detection Device,” Philips Lifeline. [Online]. Available: https://www.lifeline.ca/en/about-us/fall-detection-technology-autoalert/. [Accessed: 17-Sep-2020].


Join us now!

Become a General Member:

Sign-Up Form
Or contact us at utbiomedesign@gmail.com

UT BIOME
University of Toronto Biomedical Engineering Design Team

Home
About Us
Projects
News & Events
Contact Us
Resources

CONTACT US

utbiomedesign@gmail.com