Innovation Spotlight - Making cardiac ultrasound measurements more consistent

How was mitral regurgitation assessed before Philips developed the software?

Before 3D Auto CFQ, mitral regurgitation was assessed through subjective visualization and the Two-Dimensional Proximal Isovelocity Surface Area (2D PISA) quantification method. That method has served clinicians well and is still widely used, but it relies on two simplifications: it takes a single freeze-frame of the heartbeat, and it assumes the gap the blood escapes through is a neat circle that stays the same size throughout.⁴ In a real heart, that gap is an irregular shape, and it can change from one instant to the next.

Clinicians told us that these existing methods could produce inconsistent results: one clinical trial found that 41% of patients enrolled as having a significant leak were later judged by an independent laboratory to have a moderate leak or milder.⁵ This inconsistency can make it difficult to get an appropriate diagnosis and choose the right treatment for cardiac patients.

What would you say is the best benefit for patients with this software?

For patients, they can rest assured knowing their clinicians have a more quantitative tool in their arsenal to more accurately assess their mitral disease. Physicians have long asked for such a tool so that they are not subjectively assessing their patients.

How does the measurement generated by the software compare with manual methods?

CFQ is still relatively new, so Philips continues to work with clinical collaborators to assess its ongoing real-world benefits and how measurements are interpreted. A major challenge we encountered while developing the software was that there were no agreed gold standard for measuring mitral regurgitation; we used cardiac MRI as the best available reference.⁵ One of our clinical collaborators published a paper showing decreased variability^* compared to standard-of-care measurements, and a 2020 study found that measurements from 3D Auto CFQ matched cardiac MRI more closely than the established method did.^5,6

Can the AI in CFQ be trusted?

I feel we’re still so early in the AI journey that no one should be blindly trusting any AI tool that is put in front of them. Dr. Akhil Narang from Northwestern Medical Center puts the idea concisely: “We need to trust, but verify”. That is, even with data and results from others, the proof is in the clinician’s hands as they use AI tools and discover how the tools fit into their daily workflow.

Where do you see AI taking cardiac ultrasound in the future?

As computing power grows, I see AI taking on more of the routine parts of diagnosis. AI is very good at recognizing patterns in data it has been trained on, so large image libraries of common conditions could help physicians rule cases in or out quickly, freeing them for the unusual or difficult cases that really need their expertise.

What was your role in the development of the software?  And what personally motivated you to be a part of the development?

I was the R&D feature lead for the project within Cardiovascular Ultrasound Business. The project appealed to me from the start because it’s a clinical issue that cardiologists deal with on a daily basis. I worked with a cross-functional team across product development, clinical specialists and other R&D engineers to assemble the application. This required countless hours of discussions and clinical evaluations with key physicians. We also set up a research study with a collaborator to validate CFQ against cardiac MRI.