Charley Taylor ’87, ’91G, ’92G

Co-founder and Chief Technology Officer of HeartFlow

Today’s non-invasive diagnostic tests provide little explanation of chest pain, leading many patients to undergo unnecessary invasive testing. In fact, more than half of patients who undergo these invasive tests have no significant blockage. Using data from a standard CT scan, the non-invasive HeartFlow Analysis creates a personalized 3D model of the coronary arteries and simulates the impact that blockages have on blood flow. HeartFlow utilizes deep learning AI algorithms and highly trained analysts to build 3D models of the blood vessels that feed the heart.

What led you to starting HeartFlow?         

Starting HeartFlow was actually the result of a chance encounter. In late 1993, while I was a Ph.D. student in engineering at Stanford, I went to a talk led by the university’s new chief of vascular surgery, Dr. Christopher Zarins (HeartFlow’s co-founder). As I heard him speak about blood flow and cardiovascular health, I realized that computer modeling techniques I had been studying for the last several years, starting from my senior year at Rensselaer, could be used to quantify blood flow in patients’ arteries. I completed my doctoral research at Stanford with Chris and Tom Hughes, Ph.D., a professor of mechanical engineering at Stanford and a leading expert in computational fluid dynamics, on the topic of computer modeling of blood flow in arteries and did the first simulation of blood flow in arteries from medical imaging data. In 1997, I became a professor in the schools of medicine and engineering at Stanford and continued developing computer simulation technology for another 10 years before Chris and I founded HeartFlow.

Please explain the technology behind HeartFlow Analysis and how it is used in health care.

The HeartFlow Analysis is a new approach to non-invasive testing for coronary artery disease (CAD), the most common form of heart disease, which is the leading cause of death and affects nearly 17 million Americans today. The challenge with CAD is that it can be difficult to diagnose, because its symptoms can range from feelings of indigestion to breathlessness—or worse, sometimes patients don’t show symptoms at all. On top of this, while other non-invasive tests are available, they are not effective in helping physicians diagnose CAD and often result in additional tests, such as an invasive cardiac angiogram. Now, more than 55% of patients sent for an invasive procedure are found not to have significant heart disease, making the invasive procedure unnecessary in hindsight.

The HeartFlow Analysis aids physicians in identifying the right treatment pathway for each patient—whether it’s through minimally invasive procedures like stenting, or alternative ways of managing their disease like medication. HeartFlow utilizes a software-as-a-service model in which computed tomography (CT) images of a patient’s heart are securely sent from a hospital to HeartFlow’s application running on the Amazon Web Services (AWS) cloud. By using AI and certified human analysts, HeartFlow produces a digital 3D model of a patient’s arteries and solves the equations of blood flow using supercomputing techniques. The end result is an easy to interpret, color-coded model that physicians can interact with (e.g., zoom in and rotate) to determine if the patient has any blockages in the arteries, and whether or not the blockages are impeding blood flow and requiring treatment.

How does HeartFlow leverage data science and machine learning?

HeartFlow utilizes a SaaS model in which computed tomography (CT) images of a patient’s heart are securely sent from a hospital through the cloud to HeartFlow. Next, we use an advanced form of AI called deep learning to analyze the CT images and build a personalized, digital model of that patient’s coronary arteries. Our team of highly trained analysts then inspects this model, making any needed edits. Once this patient-specific model is completed, the HeartFlow process applies physiologic principles and computational fluid dynamics to compute the blood flow and FFRCT values at every point in the model. The completed HeartFlow Analysis is a color-coded, digital 3D model of the coronary arteries that a physician can use to determine the best treatment path for the patient.

What makes deep learning so powerful is the fact that as the algorithms are trained on more data, the performance of the product improves. HeartFlow’s deep learning algorithms have been trained using tens of thousands of CT images, and our data set continues to grow rapidly, which can lead to new population-based insights.

Additionally, we are using AI to learn which factors are most predictive of plaque rupture and heart attacks. This is really exciting and could one day remove heart-attack from its position as the No. 1 killer of men and women.

Can using the HeartFlow Analysis prevent unnecessary procedures?

Yes. By providing information on both anatomy and function, the HeartFlow Analysis provides physicians with actionable information which enables them to provide the best treatment recommendation for their patients. The technology’s real-world applicability has been demonstrated in recent studies. For example, research from the ADVANCE Registry looked at more than 5,000 patients across Europe, Japan, and North America. Using the HeartFlow Analysis allowed doctors to change their recommended treatment plan in two-thirds of patients. This meant that some patients who were due to receive stenting or a bypass were able to be treated with medication instead. Meanwhile, others who were due to be treated with medication only were identified as needing invasive management to optimize blood flow to the heart.

What is the potential for using artificial intelligence in health care going forward?

There is no doubt in my mind that AI will be one of the most important technologies in precision medicine initiatives and will be prioritized in modern health-care budgets. The possibility of improved diagnosis of disease and targeted therapies will drive this change for the foreseeable future. Leveraging AI to optimize the treatment of individual patients, whether through drugs or devices, could enable health-care businesses to provide information to physicians and hospital systems that would be priced for performance. For example, patients that might be most likely to respond favorably to a given medication or device might be identified and then this information could be provided to their physician.

 For cardiology in particular, I believe AI’s capabilities will help physicians collate data, as more of it is gathered and analyzed with deep learning technology. AI has the potential to become a powerful partner for physicians and go beyond simply aiding in the accuracy of diagnoses but help them achieve a deeper understanding of the severity of a condition and better explain to patients their symptoms and provide them with personalized treatment plans.

All that said, I don’t believe AI will ever replace physicians, but that physicians who use AI tools will replace those who don’t. am a big believer in augmented intelligence to combine the best of machine and human intelligence.

What are the biggest challenges facing health care and diagnostics today?

I think the biggest challenge is improving health-care provider awareness and overcoming the health-care industry’s capacity to adopt digital technology from a budget, bandwidth, and expertise perspective. The vast amount of technology innovations that have flooded the industry can be overwhelming and confusing for physicians and providers, so this, paired with the health-care industry’s slow-moving pace can present itself as a major challenge.

HeartFlow aims to address these challenges and ease the physician burden by changing the way cardiovascular disease is diagnosed and treated, starting with the clearest picture of a heart’s health, representing a paradigm shift in patient care by improving the overall experience and quality of life.

Health care can be a financial drain for both patients and hospitals, but with HeartFlow, physicians can become confident that they are selecting the best and most efficient care pathway for their patients. For example, the HeartFlow Analysis has been shown to reduce the number of unnecessary tests and procedures as well as reduce the overall cost of care by more than $4,000 per patient after one year. Given the fact that there are millions of patients with significant heart disease, this could save billions of dollars per year to health-care systems globally. To date, over 30,000 patients worldwide have received the HeartFlow Analysis.