AI in Mammography: A Helpful Tool, Not a Replacement for Your Doctor

by Erin Steigleder, MSW, Director of Programs

What is happening with the use of AI and mammograms?

You have probably seen some headlines over the past year or two about how medicine can make use of artificial intelligence (AI) technology, and you may have even seen something about AI being used to “look” at mammograms. So how is AI being used in mammography, and what does it mean for patients? If you’re nervous, we’ll let you know right now that radiologists are still critical to the cancer diagnostic process and cannot be replaced by AI. 

How is AI used in mammography?

AI is currently being used occasionally to help the radiologist evaluate an image to determine whether a mass is potentially cancerous or not. How does this work exactly? The AI programs, which were built using millions of mammography images, can highlight areas on a mammogram that look suspicious according to their programming. This can be beneficial by making it easier for a radiologist to catch something small or something that might be hidden in dense breast tissue. Radiologists examine images for hours, and they are only human, so their eyes get tired. Having a program that can highlight potential areas of concern can allow radiologists to spend more time and effort exercising their professional judgement based on at least a decade of education and experience, something a computer program is not able to do. 

How is this being used in Virginia?

Only a few facilities are currently using any type of AI in their mammography, but as the technology develops it is likely that more will start to use it. Some facilities will charge patients an additional fee to have their images read with AI in addition to the radiologist, but many, including facilities in Virginia, include the use of AI with no additional charge. As of right now, there is no code that facilities can use in order to charge insurance companies for the use of AI, so that’s why we have different facilities handling it in different ways. Those who do not charge for the service will cite equity as their reason for not doing so: they don’t want a new technology to benefit only those who can afford to pay for it. However, if a billing code is created for the use of AI technology in mammography, more facilities will likely be billing for it.

Not all sunshine and rainbows

Though AI can be a helpful tool in the analysis of mammograms, that is all it is: a tool. Tools are made and used by people, and because of that they are not infallible. AI has to be “trained” in order to build its programming, and it gets trained by scanning millions of images. So the programming of an AI is only as good as the images it was trained with. Some common pitfalls with this process are ensuring the racial diversity of images used to train AI. For many years, white women were more likely to get mammograms than Black women, and though that has evened out in recent years (at least in Virginia) the scientific catalog of mammography images is still mostly white. That is if the racial data is available at all for each image that is used. We know that Black women are more likely to be diagnosed younger than white women and are more likely to have dense breast tissue as they age, so it is important that AI is trained on different images from people of various racial and ethnic backgrounds. 

Another thing for us to remember when it comes to AI in healthcare is that we should never remove the human doctor from the equation. Just like spellcheck can’t write a novel, AI shouldn’t be left to make diagnoses or treatment decisions without the input of a doctor’s medical judgment. AI doesn’t actually know any more than humans do, it can just calculate things faster and highlight patterns. That is the science part of medicine. 

Doctors are vital in exercising the art of medicine: making clinical judgements, knowing when they need to gather more information in order to make a decision on a diagnosis or a course of treatment, and yes, sometimes even just following that hunch with a patient who is fine on paper but has a feeling that something is going on. For example, AI might flag a spot on a mammogram, but the radiologist knows from taking the patient’s history that they had a previous surgery in that area, and that spot is likely scar tissue. Or, AI might determine the scan is clear, but the radiologist sees something that is “funny looking” on a high-risk patient’s scan, and it reminds them of a previous patient’s cancer, so they decide to biopsy the spot anyway.

The use of AI in mammography brings us the best of both worlds, AI programming to do the work that computers do best and a doctor to do the work that human brains do best.

To learn more, read Breastcancer.org’s analysis of AI and mammography.

 

Leave a Reply

Your email address will not be published. Required fields are marked *