In a dark room at Bax Kiskun District Hospital on the outskirts of Budapest, Dr. Eva Ambrozai, a radiologist with more than 20 years of experience, stares at a computer monitor showing a patient’s mammogram.
Two radiologists had previously said that X-rays showed no signs of breast cancer in the patient. But Dr. Ambrozai took a close look at some areas of the scan, circled in red, that the artificial intelligence software flagged as possible cancer.
“This is something,” she said. She immediately ordered the woman recalled for a biopsy, which will be done within the next week.
Advances in AI are beginning to bring breakthroughs in breast cancer screening by detecting signs that doctors miss. According to early results and radiologists, so far the technology has shown an impressive ability to detect cancers at least as well as human radiologists, and this shows how AI can improve public health. It’s one of the most tangible signs to date that you can improve.
Hungary has a well-developed breast cancer screening program, making it one of the largest testing grounds for this technology in real patients. Starting in 2021, his five hospitals and clinics, which perform more than 35,000 tests a year, will be equipped with AI systems to check for signs of cancer that radiologists may have missed. Clinics and hospitals in the US, UK and European Union are also beginning to provide tests and data to help develop the system.
The use of AI is expanding as this technology becomes central to the Silicon Valley boom, and the release of chatbots like ChatGPT shows how AI has an amazing ability to communicate in human-like prose. , sometimes with alarming consequences. Breast cancer screening technology, built in a format similar to that used in chatbots modeled after the human brain, shows another way AI is permeating everyday life.
Doctors and AI developers say there are still many hurdles to widespread use of cancer-detection technology. Beyond the limited locations currently using this technology, additional clinical trials are needed for this system to become more widely adopted as an automated 2nd or 3rd leader in breast cancer screening. The tool also needs to prove that it can produce accurate results for women of all ages, ethnicities and body types. And the technology needs to prove it can recognize more complex forms of breast cancer and reduce false positives that are not cancer, the radiologists said.
AI tools have also sparked debate about whether they will replace human radiologists, with the makers of the technology facing regulatory scrutiny and resistance from some doctors and medical institutions. So far, these concerns seem exaggerated, with many experts saying the technology will only work and be trusted by patients when used in collaboration with trained physicians. there is
And ultimately, AI has the potential to save lives, said Dr. Laszlo Tabarr, a leading European mammography educator, and after examining the performance of AI in breast cancer screening from multiple vendors, the technology He said he was fascinated.
“I dream of the day when women go to breast cancer centers and ask, ‘Do you have AI?'” he said.
Hundreds of images per day
In 2016, Jeff Hinton, one of the world’s leading AI researchers, argued that the technology would destroy the skills of radiologists within five years.
“If you work as a radiologist, I think you’re like the Wile E. Coyote in the cartoon,” he told the New Yorker in 2017. under. There is no ground below this. ”
Hinton and two University of Toronto students built an image recognition system that can accurately identify common objects such as flowers, dogs, and cars. The technology at the heart of their system, called a neural network, is modeled after how the human brain processes information from various sources. This is used to identify people and animals in images posted to apps like Google Photos, and allows Siri and Alexa to recognize what people say. Neural networks have also driven a new wave of chatbots like ChatGPT.
Many AI evangelists believed that such technology could be easily applied to detect diseases like breast cancer on mammograms. According to the World Health Organization, 2.3 million people were diagnosed with breast cancer in 2020 and 685,000 died from it.
But not everyone felt that changing radiologists would be as easy as Hinton predicted. Computer scientist Peter, who co-founded Keiron Medical Technologies, a software company that develops AI tools to help radiologists detect early signs of cancer, said he knew the reality would be more complicated. I knew
Keckemesy grew up in Hungary and spent time in one of Budapest’s largest hospitals. His mother was a radiologist, so he witnessed the difficulty of finding small malignancies in his images. Radiologists often spend hours each day in a dark room looking at hundreds of images before making life-changing decisions for their patients.
“Small lesions are easy to miss,” says Dr. Edith Karpati, Kechkemesee’s mother and current director of medical products at Chiron. “It’s impossible to stay focused.”
Kechkemesee, along with Chiron co-founder and machine learning expert Tobias Raiken, said AI should help doctors. To train the AI system, he collected more than 5 million historical mammogram images of patients whose diagnosis was already known, provided by clinics in Hungary and Argentina and academic institutions such as Emory University. bottom. The London-based company also sent 12 radiologists to label images using special software that teaches AI to spot cancer growth by shape, density, location and other factors. paying a reward.
This technology creates mathematical representations of normal and cancerous mammograms from the millions of cases fed into the system. We can see each image in ways that are finer than the human eye can, so we compare that baseline to find abnormalities in each mammogram.
Last year, after testing more than 275,000 breast cancer cases, Kheiron reported that its AI software, when acting as a second leader in mammography scans, performed on par with human radiologists. bottom. It also reduced the radiologist’s workload by at least 30% because fewer of his X-rays needed to be read. In another result from a Hungarian clinic last year, he improved his cancer detection rate by 13% as the technology identified more malignancies.
Dr. Tavar’s mammogram reading technique is commonly used by radiologists, and in 2021 he tried the software and found that radiologists missed signs of developing cancer in the most difficult of his career. We retrieved some of the cases. In every case, the AI found it.
“I was blown away by how good it was,” said Dr. Tabar. He said he had no financial ties to Chiron when he first tested the technology, but has since received consulting fees as feedback to improve the system. He said he also tested systems from other AI companies, including South Korea’s Lunit Insight and Germany’s Vara, with promising detection results.
Proof in Hungary
Kheiron’s technology will first be used on patients in 2021 at a small clinic in Budapest called MaMMa Klinika. After the mammogram is completed, two radiologists will examine you for signs of cancer. The AI then either agrees with the doctor’s opinion or flags the areas for re-examination.
Since 2021, 22 AI-identified cases of cancer missed by radiologists have been documented at five MaMMa Klinika facilities in Hungary, with about 40 more under investigation.
“This is a big step forward,” said Dr. András Badassi, director of MaMMa Klinika, who was introduced to Chiron through Mr. Keckemesy’s mother, Dr. Karpati. “If this process saves one or two of her lives, it will be worth it.”
Chiron said the technique works best with a doctor, not for him. Scotland’s National Health Service will use it as an additional leader in mammography scans at six facilities, and will be introduced to around 30 breast cancer screening facilities operated by the National Health Service in England by the end of the year. Finland’s University Hospital of Oulu also plans to use the technology, and this year a bus will tour across Oman to conduct breast cancer screenings using AI.
“AI + doctors should only replace doctors, but AI shouldn’t replace doctors,” said Kechkemesee.
The National Cancer Institute estimates that about 20% of breast cancers are missed by mammograms.
Dr. Constance Lehmann, a professor of radiology at Harvard Medical School and an expert in breast imaging at Massachusetts General Hospital, urged doctors to keep an open mind.
“We are not unrelated,” she said. “But there are some jobs that can be done better with a computer.”
Dr. Ambrozai of Bax Kiskun District Hospital outside Budapest said he was initially skeptical about the technology but was quickly persuaded. She took out an x-ray of her, a 58-year-old woman with a small tumor discovered by AI, which Dr. Ambrozai had difficulty finding.
The AI saw “something that seemed to come out of nowhere,” she said.
