By Jonathan Woods

A.I. Vali’s five-year journey from idea to clinical trial for AIDREA, its AI-powered cancer detection device, required the Toronto-based startup to develop a new technology built to meet the strict technical and regulatory standards in health – especially around patient privacy.

“Ultimately, our purpose is to help the clinician improve their accuracy of diagnosis as early as possible so patients can get the treatment as fast as possible,” says Azar Azad, A.I. Vali’s co-founder and CEO.

But training models in the health space is a challenge. Training requires large and varied datasets from multiple institutions, but hospitals are reluctant to share patient data, as virtually any risk that may lead to its compromise is considered unacceptable.

Solving this puzzle – getting sufficient data without needing hospitals to share it – required an advanced AI solution. To find it, A.I. Vali turned to collaboration with Vector.

AIDREA uses computer vision to provide real-time video image analysis during endoscopies, a procedure that involves inserting a flexible, camera-mounted tube down a patient’s esophagus to give doctors a look inside the throat. The goal is “to support clinicians improve the accuracy in early detection of diseases, such as cancer while improving the workflow of the clinics and reducing the patient waiting time,” Azad says.

She previously held a post as Director of Research Services & Personalized Medicine before striking out on the entrepreneurial path, with a focus on medical devices, in order to improve patient care. Azad’s background gave her fluency with clinical research and an ability to identify opportunities to leverage technology. What Azad required to make AIDREA a reality were experts with a roadmap through the state-of-the-art in AI. Her idea was incubated in NEXT AI and then the Creative Destruction Lab, where one of its program advisors, Graham Taylor, who happened to also be a faculty member and project manager at Vector, suggested the company get involved with Vector.

As a participant in Vector’s FastLane Program for small and medium-sized companies currently using AI or primed to do so, Azad learned about Vector’s Privacy Enhancing Techniques (PETs) bootcamp. An intensive program focused on cutting-edge methods for training models while adhering to strict privacy requirements, the bootcamp offers tutorials on advanced privacy techniques and a fully-guided technical onboarding that set up resources for A.I. Vali’s team. Then, over three days, participants use the techniques and resources to build a prototype for one of their own real business use cases.

For A.I. Vali, the very first tutorial was a revelation.

Azad and A.I. Vali machine learning engineer Rohini Gaikar were introduced to horizontal federated learning, a technique that can allow several organizations to collaboratively train a single model without ever having to share their data. Here’s how it works in health: each hospital uses its data to train a model on its own. Those trained models are then sent to a central server, which aggregates them and updates the parameters of a global model. That global model is then sent back to each hospital for further refinement through more training. Over many such cycles, the global model achieves the required standard of accuracy – and, most importantly, does so without any patient data ever leaving any hospital’s own secure infrastructure.

“To convince a hospital to participate in a trial, I have to communicate that you’re involving AI, but without any risk to your data confidentiality,” Azad says. “After we heard about federated learning, we talked with two hospitals, explained at a high level what we wanted to do, and got their agreement to participate in our trial.”

“For this, I give significant credit to Vector,” she says.

The technique is now being embedded in AIDREA as the company makes a push for a trifecta of health milestones: ISO certification, Health Canada approval, and an FDA clinical trial scheduled for early 2023.

For Azad, even more exciting than being on the cusp of commercialization is the possibility of improving the work of clinicians and the health of patients. “That has an impact on the entire health economy – for the hospital, the health system, for the patient’s quality of life, and for the cost of patient management.”