SEEING THE BIG PICTURE: USING ARTIFICIAL INTELLIGENCE TO RECONSTRUCT THE 12-LEAD ECG (CCC-210)

What problem does your idea solve?: The gold standard for ECG interpretation uses a 12-lead ECG. However, interpretation of the 12-lead ECG can be often limited due to noise or missing data in particular leads. For example, this can impact the identification of cardiac structural abnormalities (i.e. acute or prior infarct, chamber enlargement, etc) and limit the diagnosis of arrhythmias. Furthermore, patients are also increasing using handheld ECG devices, that only provide for limited leads (often only a single-lead or up to 6 limb leads). Our Cardiovation seeks to use artificial intelligence to reconstruc the 12-lead ECG using reduced ECG data. In cases where ECG data may be partially missing or incomplete, this Cardiovation has the opportunity to substantially improve our diagnostic abilities.

Objective: Using artificial intelligence to reconstruct the 12-lead ECG in cases of missing ECG data or limited lead data.

At what stage of development are you? Have you tested your innovation? : We have worked with engineering science students from the University of Toronto and partnered with the Department of Computer Science at the University of Toronto to develop a model that is trained using publicly available ECG datasets. These publicly available datasets (PTB-XL, contain approximately 22,000 ECGs for 19,000 patients) and are used the train and internally validate our ECG reconstruction model. Specifically, we evaluate 7 different models (including linear regression ranging to convolutional neural networks) as a means to reconstruct the 12-elad ECG using resticted leads. We identified the highest performance in ECG reconstruction could be achieved using 3-leads (lead I, lead II, and precordial lead V3). Using these 3-leads, we were able to reconstruct the full 12-lead ECG with high fidelity. Our current next step is external validation of the 12-lead ECG reconstruction, using clinical ECGs from a dataset at Sunnybrook hospital.

Explain how and where this impacts the cardiovascular care patient/workplace.: The 12-lead ECG is fundamental for cardiac care. It is fundamental for the detection and recognition of ischemia and arrhythmias, and cardiac structural abnormalities, including chamber hypertrophy or infarction. However, 12-lead ECGs can be recorded with variable quality, often due to lead instability, breathing or motion artifact, or other patient-specific factors (i.e. chest tubes and/or dressings, LVAD, etc). In these cases, detection of cardiac abnormalities can be limited due to the reduced quality or leads available for review.

Reconstruction of the 12-lead ECG in cases of missing ECG data is of high importance, as it may generate actionable ECG data that may otherwise be missed. This Cardiovation seeks to employ artificial intelligence to reconstruct missing ECG data based on leads that are already available..

Why do you think your innovation should receive the $1,000 prize? : We believe this innovation should receive the $1000 prize because it delivers potentially actionable clinical data that physicians can interpret. Cardiologists are already comfortable with 12-lead ECG, but are often limited in their interpretation due to baseline artifact (wander), noise, or missing ECG leads. This can impair the diagnosis of cardiac structural abnormalities, ischemia, or arrhythmias. However, with reconstruction of the 12-lead ECG to overcome this missing data, we are able to provide the cardiologist with full 12-lead data for interpretation. Our innovation is well-suited for the Cardiovation challenge, because we use already available data to create an innovation that can have improve the cardiac care that we are already delivering. By leveraging artificial intelligence, we are able to fill in gaps in ECG data and facilitate routine cardiovascular care.

If you should be the winner, what will you use the prize money for?: If we win the Cardiovation competition, the prize money will be used to support ongoing research with ECG reconstruction. Our current work is under consideration at Sensors (open access journal), and support from the competition will be used to support research and ultimate publication of this important work. Our code will be made freely available (open access), and there is an associated fee with open access publication of this work. The prize money from the Cardiovation competition will be used to support this.

Provide a brief overview of your pitch. How are you going to sell this idea to the dragons and the audience?: For the Cardiologist, the 12-lead ECG is fundamental to our practice. Some may even consider as the 5th vital sign - our clinical practice requires both rapid and accurate ECG interpretation. But our ECG interpretation can sometimes be limited by factors beyond our control - artifact, noise, patient movement, leads falling off - all examples of Incomplete Data. Incomplete data - in this case, an incomplete ECG - substantially slows and impact our ability to interpret the 12-lead ECG. Is that noise or real QRS complexes that we are seeing? Is there ST elevation, or just a wandering baseline?

With our Cardiovation, we seek to overcome these limitations with artificial intelligence. Even when leads are missing, we can reconstruct the missing data from the other available leads. Our machine learning algorithms reconstruct a clean and interpretable 12-lead ECG, so that YOU can get back to focusing on the patient!