We are a quality improvement lab for neurovascular care. We coined stroke resuscitation, built the Protected Code Stroke protocol, and now use machine learning on bedside signals — including voice, in Sonic Diagnosis — to make acute stroke care faster, safer, and more humane.
Lab PI: Dr. Houman Khosravani, Division of Neurology, University of Toronto (Sunnybrook Research Institute).
Built on the WICG HTML-in-Canvas proposal (drawElementImage). In browsers without the flag, an equivalent Canvas 2D rendering is shown. How this site is built.
We termed this work “stroke resuscitation”: applying medical and trauma resuscitation principles — and Crisis Resource Management (CRM) — to the hyperacute code stroke, with an active simulation program aimed at metrics like door-to-needle time. This includes our work within NICE (Neurovascular Innovations CollaborativE).
#AI4QI / #ML4QI — machine learning to bolster current standards of inpatient acute stroke care, applied to datasets and time-series data. Our flagship, Project MASA (Machine Learning Assisted Swallowing Assessment), explores voice as a biomarker: Sonic Diagnosis.
Maximal effective care includes palliative care. We develop protocols for its routine integration into stroke treatment and define the scope of palliative medicine in acute stroke — compassionate, patient-centred care across the spectrum, alongside aggressive intervention when appropriate.
Lab members also produce Stroke FM, the official podcast of the Canadian Stroke Consortium.
An illustrative snapshot of emphasis across current projects — a guide to the lab’s centre of gravity, not a measured dataset. In supporting browsers, the chart labels are real HTML elements drawn into canvas with drawElementImage (try tabbing to a slice); otherwise the same labels are rendered as canvas text.
Our lab championed crisis resource management in stroke simulation to optimize critical intervention metrics such as door-to-needle times, and published the first reframe of CRM for stroke care. During COVID-19 we developed the Protected Code Stroke protocol — integrated into national and international guidelines and downloaded 47,945 times from the American Heart Association’s Stroke journal site, with 171 citations (publisher metrics, August 2026).
Simulation here is a QI instrument, and it moves real numbers: alongside PDSA-cycle redesign, Sunnybrook’s median door-to-needle time fell 25%, from 38.5 to 29 minutes — below the national average and durable across a service managing more than 1,000 code strokes a year. Sunnybrook ranks in Ontario’s top two for low 30-day stroke mortality and readmission.
We call the discipline neurovascular resuscitation: care pathways and human performance, rehearsed until the team moves as one. Within NICE, our lab leads the medical / neurocritical care side of hyperacute care, while Dr. Hawkes helms the neurovascular and catheter-based techniques.
Each year the lab typically takes one student through the T-CAIREM summer studentship, and in some years a student via SRI or CREMS. If you work on QI, human factors, simulation, or clinical machine learning — as an engineering, CS, or MD student — we would like to hear from you.
A zero-dependency static site — no framework, no build step, no trackers, no cookies — served with a strict Content-Security-Policy. Charts and the hero diagram use the experimental WICG HTML-in-Canvas API where available, with an accessible HTML equivalent always present. Enable chrome://flags/#canvas-draw-element in Chrome Canary to see live HTML rendered inside canvas.
For more news and updates, visit strokeinnovationlab.ca.