Prof Nobuyuki Kagiyama

I'm Dr Kagiyama from Juntendo University, Japan, and I work as a clinical doctor there. I do echo mostly, I'm kind of echo guy, but my research interest is in artificial intelligence and the implementation of the AI technology in real-world clinical practice.

Could you tell us about the algorithm, study design and patient population behind this trial?

In this trial we used a platform called Us2.ai, which combines two algorithms. One is a parameter-based one and this is the software: AI automatically measures all routine echo parameters, and based on those measured parameters, the AI judges whether the patient is likely to be amyloidosis.

And the second one is a deep-learning model. This model looks at the echo four-chamber view, and from the four-chamber senior loop, the deep-learning model evaluates the probability of having cardiac amyloidosis.

This was a retrospective, real-world analysis of the all-consecutive echocardiography performed at Juntendo University during one year. So the patients are real-world, consecutive patients who underwent echocardiography for clinical indication.

What were the key findings?

So the key finding is that this AI software plugged or identified about 5% of the entire [illegible] cohort as potentially having cardiac amyloidosis. Among patients who underwent clinical diagnostic testing afterwards. Now AI-positive patients demonstrated about 17% of real cardiac amyloidosis, whereas the AI-negative patients essentially had no amyloidosis.

What are the take-home messages for practice?

So I see this as a practical enrichment of the probability of cardiac amyloidosis. I don't think this will be a standalone test for detecting real cardiac amyloidosis, but this AI technology will pick up those patients who are more likely to be amyloidosis, and you can proceed with other advanced imaging techniques afterwards.

What are the next steps for this technology, and how do you see AI transforming the cardiovascular imaging and disease screening over the next few years?

So these kinds of AI technologies will make imaging a more objective and more automated routine process of identifying hidden diseases, cardiovascular diseases. Next step we will perform like prospective trials or prospective studies where we really test all patients flagged by the AI. So with those kind of studies we will be really able to evaluate the real performance of AI.