On the TechTO stage ×1
First seen on the TechTO stage in 2018. Every talk is searchable — ask the archive about Jay ↗
In their words
When you're developing machine learning modules for healthcare, you have to have real patient datasets.
We built a hospital room, hired actors, had scripts — then from there we started getting connections to the actual patient rooms.
The nursing staff tells us, in these very busy environments, if there's too many false positives they'll ignore the system. There's too many alarms going on already.
Quick answers
Who is Jay Couse?
A co-founder of Spectrum AI. TechTO's host introduced him at HealthTO in October 2018 as a serial inventor and entrepreneur with five startups and 82 patents filed, and on stage he described founding Spectrum AI with Dr. Chance, an orthopedic surgeon. No current role is published here because none could be verified from a first-party source.
What was Spectrum AI building?
Computer vision, built on deep learning models, to recognise falls among high-risk patients in places where nobody is watching, starting with long-term care. The system used a fixed camera mounted about nine feet up in the corner of each room rather than anything the patient wears. As presented at HealthTO, October 2018.
Why did he start it?
He said the idea came from a conversation with his co-founder about their mothers-in-law, both in long-term care homes and both of whom had been hurt. He described one being found on a shower floor after several hours, only because water was leaking into the hallway, and said about 75% of these falls happen unwitnessed in a resident's private room.
How did Spectrum AI build its training data?
It started from public datasets, then built a mock hospital room and filmed actors working from scripts, then moved to cameras in real patient rooms and added synthetic data, including augmentation that turns a single recorded action into thousands of similar ones. West Park Healthcare Centre acted as the live test site.