Deep Learning Algorithm to Diagnose Cardiovascular Risk Factors

Google Research released a blog post last Monday regarding their recent headway with a deep learning algorithm helping to assess cardiovascular (CV) risk factors.

Previously, retinal images had been used to increase the accuracy of diagnosing diabetic eye diseases. Researchers at Google are using similar retinal images to accurately predict other indicators of CV health.

The algorithm in question was trained on data from 284,335 patients and can discern retinal images belonging to smokers vs. non-smokers 71% of the time.

The algorithm was also able to accurately predict 70% of the time, given their retinal scans up to five years before the event, which patients would go on to suffer a CV event such as a heart attack or stroke.

The CV risk factor heavily increases when multiple indicators are at play, including:

• genetics (such as age, gender, other illnesses)
• lifestyle components (such as blood pressure, cholesterol, physical activity, and smoking and drinking habits)

Currently, disease diagnoses are based on guesses and tests.

While some factors can be disclosed by the patient directly, others like cholesterol levels require a blood draw—something not everyone is comfortable with.

In addition, the doctor may also ask for an echocardiogram and/or electrocardiogram to further narrow down or rule out CV risks entirely. Time to diagnosis can often be prolonged if symptoms don’t fit the general bill of CV risk factors.

According to the World Heart Federation, there are 17 million deaths related to CV diseases per year, and 80% of those could be preventable.

 

Google Research’s work on this deep learning algorithm could pave the future towards cutting down on guesswork and focusing further on confirming the diagnosis and immediate treatment options to prevent a CV event.

We look forward to hearing positive results from this algorithm’s performance in catching these preventable deaths.