SickKids gets ready to deploy their new AI apps developed in-house

Toronto’s Hospital for Sick Children (SickKids) is poised to confirm its leadership in AI research with the introduction of several machine learning models for patient care. “We’re at the point now where we’re going to start turning on a bunch of solutions over the next year,” said Robert Greer, technical director of the SickKids AI initiative. “People are going to know about SickKids and AI in a pretty big way.”

Patients seeking care at SickKids’ Emergency Department (ED) will be among the first to benefit from a series of machine learning algorithms developed in partnership with SickKids ED physician Dr. Devin Singh, co-lead of SickKids AI Service and co-founder and CEO of Hero AI. These models leverage information from the triage process to help streamline care and reduce wait times.

If a patient exhibits symptoms characteristic of appendicitis, for example, one model will expedite ordering of diagnostic tests right away, a huge improvement over the current workflow that requires a child to wait, potentially for hours, to see a physician and only then be sent for blood tests and imaging.

Using the model to predict the likelihood of appendicitis, ED physicians will have test results when they initially see their patients, streamlining decision making and treatments. Similarly, if information conveyed in the triage encounter is indicative of a mental health crisis, an AI model currently deployed helps to reduce wait times by automatically alerting the psychiatric team to attend to the patient.

Another ED triage model will soon be deployed to alert clinicians to expedite interventions and administer medication if the wait time for a child with sickle cell disease in pain exceeds an acceptable threshold.

Also aimed at streamlining operations is an ED census model that predicts patient numbers by hour based on six years of historical data. If the model predicts a surge in patient numbers, ED management can open up more space and bring in more staff. “On the flip side,” said Greer, “we can close areas and send some of our staff to other areas of the hospital if the model predicts fewer patients.”

Most of these ED triage models are currently in a silent trial phase awaiting final approval but should be in clinical use by early next year, said Greer, who points out that the application of AI in the ED is just the tip of the iceberg.

One model in development for the ICU uses waveform data from medical devices to alert clinicians if a patient’s condition is worsening. Another model in the imaging space detects hydronephrosis, a kidney disorder, to speed a decision for surgery.

SickKids is “somewhat conservative” about deploying machine learning models into clinical practice because “at the end of the day patient safety is our top priority,” said Greer, noting that in an era where the regulatory landscape is actively changing, SickKids has developed its own policies and standards.

“One of the biggest challenges we face as an organization is that there’s little government funding for AI work,” said Greer. “It’s largely funded by grants and we’re lucky to have a great foundation that works to raise money and support work like this.”

One important donor is Pure Storage, which gifted high performance storage equipment to SickKids several years ago.

“In order to train a machine learning model and predict the future, you need a lot of historical data, and you need a place to put it,” explained Greer. “It’s a very computationally and data intensive activity, so it’s important that the storage we have is as high performance as possible.

“The vast majority of machine learning models that we train and build to solve a clinical problem will fail, so we need an infrastructure solution that allows us to fail as quickly as possible because eventually, one will work, and the faster we get there, the faster we can implement a solution and have an impact on care.”

AI adoption is rapidly increasing across healthcare and life sciences (HCLS), but many organizations are struggling with the infrastructure needed to meet the high-performance data and energy demands. This partnership is driving an end-to-end transformation in healthcare and life sciences, providing exceptional performance and reliability to meet the growing demands of modern AI applications.

“Pure Storage FlashBlade is a game-changing solution, purpose-built for modern, unstructured workloads. It simplifies data management and accelerates AI processes, significantly boosting productivity for data scientists and agility for data architects,” said Bill Bryer, the company’s healthcare lead for Canada.

“Pure Storage offers a centralized, efficient data platform integral to deep learning architectures. This powerful combination enhances the productivity of data scientists and simplifies scaling and operations,” explained Bryer.

The synergy between pure storage’s massively parallel storage architecture and AI ecosystem partners accelerated computing is transforming key HCLS use cases such as medical imaging, clinical decision support and pathology, genomics and drug discovery.

“Pure Storage’s equipment was instrumental in getting us to where we are today, but we’re at the point now where we’re going to have to dramatically scale up what we have,” said Greer.

Fortunately, Bryer is a big fan of SickKids’ AI work and is excited to expand the partnership to ensure the hospital’s infrastructure remains fast and easy to upgrade as they continue building out innovative new means of patient care.

State-of-the-art storage and processing infrastructure will also help SickKids as it engages with other paediatric hospitals in Canada and the U.S. and explores opportunities to market its expertise.