Dr. Amy Patel is the Medical Director of the Breast Cancer Center at Liberty Hospital outside Kansas City.
She spoke about an AI algorithm they use with breast ultrasounds that gives radiologists a data-informed second opinion.
For the month of October, we’re focusing on the intersection of healthcare and artificial intelligence.
Dr. Amy Patel: The way that we are utilizing it — and many practices and institutions to utilize this algorithm, how they're using it is — let's say that a patient comes in and they have a mammogram and they're maybe we see something,
And let's say we call them back for additional testing, and we might do additional mammographic views and an ultrasound, and then on that ultrasound we may see something there.
When we see something there, we can decide, do we want to use the AI algorithm or not? So, you don't use it on every patient. It's just on select cases.
So, basically, the tool will analyze the lesion, and it analyzes the lesion in just seconds. So, basically, it's a machine learning algorithm, and it's based on 2 million cases with both radiologic and pathologic correlation. It analyzes a lesion based on 17,000 features of interest. So, it's pretty incredible.
Basically, once you analyze the lesion, it essentially renders a risk assessment, you know, is this benign? Probably benign? Is this something that may be suspicious and needs biopsy?
So, ultimately, it acts as a second opinion consult, or clinical decision making. So, at the end of the day, the radiologist has to make the call of what to do with the patient.
It's all a very secure network — there's no identifiable patient information that's going anywhere, but it is, you know, these cases are being grouped with cases that are being used where other institutions and practices are using this tool, now, in over 65 countries. I mean, it's pretty incredible.
This tool has been FDA approved, and so, it's not some sort of experimental algorithm.
Over time, we've seen improvement of, you know, just utilizing the tool more frequently, and, you know, I've never seen — knock on wood — I've never seen this algorithm under call anything, like, say that something is benign and it was a cancer.
I have seen over calls or false positives — especially in the beginning when we were using the tool, but over time with consistent use, I have significantly seen a drop off.
So, what we've been able to see, now, is we've been able to maintain our cancer detection rate, but reduce our unnecessary biopsies or false positives by almost 40%. So, that's pretty significant.
So, we talk about patients and cancer care and detection and about, you know, mortality — we want to reduce breast cancer deaths, but also, we need to be thinking about morbidity.
And morbidity of a patient happens when you introduce a needle in a patient's breast, and there's always a risk with that, you know, whether it's infection, a blood collection called a hematoma.
And so, if we can also mitigate that as well — and just the cost of a biopsy, I mean, they're not cheap. Image guided biopsies are very expensive. So, if we can't eliminate the cost for the patient as well, it's truly a win-win.
