Dr. Fu Siong Ng is fed up with one-size-fits-all medicine. As a cardiologist who subspecializes in electrophysiology – the electrical properties of the heart – he has worked for decades to treat patients with potentially life-threatening conditions. These interventions can be transformative, but often they are untailored. Given the huge variation possible in how conditions may show up in patients, that’s a big problem.
To take one example, atrial fibrillation is an irregular and often very rapid heart rhythm that can lead to blood clots in the heart, and those clots can create other problems such as strokes. Just in the US, the condition affects 1 – 2% of the population – millions of patients – and costs health insurers over $25 billion a year. One leading treatment is to kill small areas of heart muscle that may lead to the condition, a procedure known as ablation. The procedure literally burns muscle. Doctors have been performing ablations for decades, often with good results. And yet the procedure is typically not tailored to the patient despite the many ways that heartbeats can be irregular. The same tissues are typically addressed, in the same ways.
Dr. Ng leads a multi-disciplinary 12-person laboratory at Imperial College in London that’s applying advanced methods of computer science, including AI, to problems in electrophysiology. When it comes to ablation, he’s inspired by an unlikely source. He told me, “Google’s Pixel phone can take out somebody from an image and replace the person with what the background probably was. What if we could apply that technology to the heart?”
Dr. Ng leads a multi-disciplinary 12-person team bringing together physicians, biologists, ... [+]
Dr. Ng explained his reasoning. “When we perform ablation today, we create a map of the heart to plan where to deliver the ablation treatment. But the electrical information from the map is very incomplete, which means we have to treat patients in a highly standardized way. To create the map, we place electrodes on the inside surface of the heart – usually around 20 of them at a time – via a catheter, to record electrical signals from the heart. Unfortunately, each electrode gives information on only a very small radius around it, so altogether, we may have a map of only 5% of the heart at a time. That’s not enough to create varied treatments.”
Working on this problem is one of Dr Ng’s PhD students, Alexander Jenkins, a physicist who joined Dr Ng’s group to apply his knowledge to important problems in cardiology. “What if,” Jenkins adds, “we could fill in the blanks, using AI technology like the Pixel’s? A generative diffusion model, which you also see used in image creation tools like DALL-E? We could train the model based on the physics of the heart and the way voltage propagates through it, enabling the AI to create a sufficiently accurate picture of the gaps to establish a tailored treatment plan, even when 95% of the heart is actually invisible to us.”
With cancer, we’ve already seen the power of personalized treatments, based on genetic mutations or even on one individual’s DNA. But these type of interventions haven’t been as possible in electrophysiology because patients couldn’t be viewed on an individual scale without extraordinarily invasive procedures that would create their own problems. As Dr. Ng puts it, “AI lets us see less and fill in the gaps.”
AI has countless applications in healthcare. It can free doctors from taking tedious notes. It can discover new drugs. In diagnostics, AI’s potential to see patients on an individual scale unlocks the ability to create new, tailored therapies and treatment plans. Oftentimes, these therapies already exist but are applied without adjustment to patients’ particular circumstances. The improved diagnostic can facilitate major leaps forward in how medicine is practiced. Dr. Ng’s and his group’s explorations may be only the beginning of what’s possible.
