Zachi Attia had never received any surgical nor medical training. He almost fainted the first time he witnessed an actual cardiac surgery. Yet, he has to spend most of his days at one of the most prestigious medical institutions in the US. Attia is a machine learning expert and he plays an important role in the country’s effort to reform heart disease treatment with the use of artificial intelligence (AI).

By working alongside with physicians and sharing an office with some of them, Attia had created algorithms that would predict whether any individual is down with hidden cardiac abnormalities, before he or she actually experiences any symptoms. However, identifying high-risks patients that are invisible even to the most experienced cardiologists is just a short-term goal. In the long run, Attia aims to integrate AI into routine medical practice.

Medicine and technology as an inseparable affair

This is what Mayo Clinic is trying to achieve at the moment: placing technology experts and medical professionals, side by side. The former shall follow the latter through rounds and observing how medical procedures are performed, to equip themselves with thorough understanding of how to render care and clinical challenges that are hindering the process. Attia is one of the five software engineers and data scientists involved in the effort. Over the past three years, they have published more than 20 studies on the use of AI in cardiology, including field testing of algorithm in primary care clinics.

Dr. Eric Topol, Cardiologist and a prominent AI figure from the Scripps Research Translation Institute in San Diego called this an “acid test”, because we cannot implant whatever that has been validated using only computers directly on patients. As such, everything has to be built within the hospital using actual and clinically meaningful patients’ data. Algorithms have to work on a diverse group of patients with complex medical issues without bias and unknown outcomes.

One of Attia’s most recent projects concerns the use of an AI-driven electrocardiograms (ECG) which allows physicians to detect early onset of atrial fibrillation via convolutional neural network. A study published last summer indicated the algorithm was 80% accurate in identifying patients with atrial fibrillation before they experience any symptoms.

In a particularly incident involving a patient who had undergone ECG over a period of 30 years but never officially diagnosed with atrial fibrillation. The AI system noticed two occasions when the patient was at high risk. The patient could have been treated with anticoagulants at least seven years beforehand to avoid these risks.

The dilemma and benefits

This is where the tricky part kicks in; having patients to start on blood thinner prior to a verified episode of atrial fibrillation means putting the patients at risks of serious bleeding and may limit their daily activities. This contradicts with the initial goal of intervening early. Thus, Attia and fellow physicians have to examine carefully how the use of algorithm may impact clinical decision making and patient outcomes in the near future. Before that, AI system will remain an advisory tool.

Nevertheless, Attia believes AI can make a different especially thousands of Americans die each year from treatable cardiac condition as most of their problems were not recognized in time for early interventions. “We want to make sure that, for anything that is treatable, we catch it as soon as we can. If we can solve it before it becomes a big deal, then we can really affect how cardiac care is delivered” he told Stat.

Besides, Attia also learnt to appreciate clinical data beyond lines of monotonous codes but insights into a person’s heart functions which may drop hints of any impending malignancy. He will now improve algorithms in a way which best fit into physicians’ practice, making them more relevant in addressing the blind spots of healthcare.


Author Bio

Hazel Tang A science writer with data background and an interest in the current affair, culture, and arts; a no-med from an (almost) all-med family. Follow on Twitter.