[ad_1]
Despite major advances in genetics and modern imaging, the diagnosis surprises most patients with breast cancer. For some, it's too late. Further diagnosis involves aggressive treatments, uncertain results and more medical costs. As a result, patient identification has been a central pillar of breast cancer research and effective early detection.
With this in mind, a team from the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) and Massachusetts General Hospital (MGH) has created a new in-depth learning model to predict from a mammogram, if a patient is as likely to develop breast cancer. like five years in the future. Trained in mammograms and known outcomes in more than 60,000 patients with MGH, the model has learned subtle patterns of breast tissue that are precursors to malignant tumors.
Professor Regina Barzilay of MIT, herself a breast cancer survivor, said it was hoped that such systems would allow physicians to personalize screening and prevention programs at the individual level, making late diagnosis a relic of the past .
Although it has been shown that mammography reduces breast cancer mortality, the frequency of screening and the timing of the beginning of the debate are the subject of debate. While the American Cancer Society recommends annual screening beginning at age 45, the US Preventative Working Group recommends screening every two years starting at age 50.
"Instead of taking a unique approach, we can customize cancer risk screening for women," said Barzilay, lead author of a new paper on the project released today. in Radiology. "For example, a doctor might recommend to a group of women to have a mammogram every two years, while another high-risk group might undergo an additional MRI. at MIT and a member of the Koch Institute for Integrated Cancer Research at MIT.
The team model was significantly better at predicting risk than existing approaches: it accurately placed 31% of all cancer patients in its highest risk category, compared to only 18% for traditional models.
Constance Lehman, a Harvard professor, said the medical community has so far had little support for risk-based rather than age-based screening strategies.
"That's because before, we did not have an accurate risk assessment tool that was tailored to every woman," says Lehman, professor of radiology at Harvard Medical School and head of the imaging division. at the MGH Hospital. "Our work is the first to show that it is possible."
Barzilay and Lehman co-authored the article with lead author Adam Yala, a PhD student at CSAIL. Tal Schuster, doctoral student, and Tally Portnoi, former master's student.
How it works
Since the first breast cancer risk model of 1989, development relies heavily on the knowledge and intuition of what could be the main risk factors, such as age, family history breast and ovarian cancer, hormonal and reproductive factors and breast density.
However, most of these markers are only weakly correlated with breast cancer. As a result, these models are still not very specific at the individual level and many organizations continue to believe that risk-based screening programs are not possible because of these limitations.
Rather than manually identify patterns of a mammogram at the origin of future cancer, the MIT / MGH team formed an in-depth learning model to infer patterns directly from the data. Using information from more than 90,000 mammograms, the model detected profiles too subtle to be detected by the human eye.
"Since the 1960s, radiologists have found that the mammary tissues seen on mammograms are very diverse in women," says Lehman. "These patterns can represent the influence of genetics, hormones, pregnancy, breastfeeding, diet, weight loss, and weight gain. We can now use this detailed information to be more precise in our risk assessment at the individual level. "
Make cancer detection fairer
The project also aims to make risk assessment more accurate for racial minorities in particular. Many early models were developed on white populations and were much less accurate for other breeds. The MIT / MGH model, on the other hand, is just as accurate for white women as it is for black women. This is especially important since black women are 42% more likely to die of breast cancer because of a wide range of factors that may include differences in screening and access to health care.
"It is particularly striking that the model works for both whites and blacks, which was not the case with the previous tools," says Allison Kurian, associate professor of medicine and health research at the University of Toronto. the Stanford University School of Medicine. "If validated and made available for widespread use, it could really improve our current risk assessment strategies."
Barzilay says their system could also someday allow doctors to use mammograms to see if patients are at greater risk of contracting other health problems, such as cardiovascular disease or others. cancers. The researchers want to apply the models to other diseases and conditions, particularly those with less effective risk models, such as pancreatic cancer.
"Our goal is to integrate these advances into the quality of care," says Yala. "By predicting who will develop cancer in the future, we can hope to save lives and fight cancer before the symptoms begin to manifest themselves."
[ad_2]
Source link