Artificial Intelligence: A new technology in mammography is not just a game changer, but a life saver
Using artificial intelligence to help read mammograms may increase cancer detection rates and reduce false positives.
Mecklenburg Radiology Associates (MRA) strives to provide the best patient care in the industry and takes pride in bringing cutting-edge techniques to the detection of breast cancer.
Our women’s imaging team is currently exploring four different FDA-approved software platforms that use artificial intelligence (AI) to help radiologists read mammograms more accurately. These novel platforms could ultimately save lives by detecting more breast cancers earlier, said Dr. Matthew Locker, a women’s imaging radiologist at MRA who is leading the effort.
Dr. Locker said using AI also has the potential to reduce the recall rate, which refers to the number of women called back for further testing, because it holds promise in more accurately identifying mammograms that are negative for breast cancer.
We asked Dr. Locker to tell us more about this advanced technology, which he estimates is currently used in fewer than 5% of hospitals in the United States. To his knowledge, the technology is not yet being used in the Charlotte region. It is still in the testing phase at MRA.
Q: What exactly is AI as it pertains to mammograms, and how does it work?
AI software is almost like having a second set of eyes that looks at mammograms behind the scenes. It can analyze both 2D digital mammograms and 3D digital breast tomosynthesis.
AI uses something called deep learning neural networks, described by experts as a system of nodes and neuronal connections, similar to a human brain. The network assigns probabilities and weighting factors to different parts of the image, some of which are undetectable to the human eye and radiologist. AI is able to make imperceptible connections to arrive at a diagnosis of breast cancer in a more accurate way, allowing even earlier detection.
Q: Would this represent an advancement on what radiologists alone can detect?
Yes. Radiologists will read tens to hundreds of thousands of mammograms in their careers. But AI has already analyzed millions of mammograms, and the information it extracts is significantly greater than that obtained by humans.
In most studies, radiologists alone and artificial intelligence alone perform fairly equivalently. However, when you add both together, radiologists get better in a tangible way.
Radiologists show improvement not only in terms of the percentage of cancers detected but also demonstrate a lower recall rate. Most mammograms we read are normal, so we do anything we can to reduce needless recalls and decrease patient anxiety. Research shows that virtually all of the core metrics we track as mammographers are improved with AI. And that ultimately equates to saving more lives.
Q: Can AI detect all types of breast cancer? What else is it better at doing?
Yes, AI can detect all types of breast cancer. One of the most difficult parts of mammography is something called interval cancer – cancer that is found after less than a year after a negative mammogram. Those typically are very aggressive, life-changing cancers.
Some studies have found that when AI is combined with other clinical data, like breast density, BMI, and age, researchers could predict 30% to 50% of interval cancers not picked up at the time of original screening mammogram. That’s staggering. That’s not only a game changer — it’s a life changer!
Another area of encouragement is in women with dense breast tissue, some of whom may not qualify for other diagnostic modalities like MRI or ultrasound. AI is able to flag patients with dense breast tissue who might benefit from these supplemental studies.
Q: How can AI help in parts of the country with a shortage of radiologists?
Right now, there is a lot of demand for radiologists, and in some rural areas and third world countries where radiology resources are limited, AI holds promise to augment care provided by existing radiologists, resulting in significant expansion of services available to patients.
Q: How soon might this technology come to the Charlotte region?
If real-world testing is successful, I think most practices will be using AI in mammograms within the next two to three years. It impacts patient care in such a profound way; we owe it to patients to make it available. As more institutions use AI and prospective studies are completed, the medical imaging community will have more confidence in it. It is very exciting.
