Throughout her career as a radiologist, Connie Lehman, MD, dedicated herself to the early detection of breast cancer. However, she came to the realization that there were limited imaging tools available for assessing the risk of breast cancer, and those that were there were heavily outdated.
“I’ve seen the power of the image to find breast cancer before it can be felt by the woman, which is a time when we can effectively cure breast cancer,” Lehman, a breast imaging specialist at Mass General Brigham, told Inside Precision Medicine. “I recognized that there was much more available than just the mammogram for women at increased risk, including the power of MRI and extensive early research and screening with MRI to help ensure that women’s cancers are identified earlier. What I realized was that we were making incredible advances in imaging technology. Still, we weren’t matching that with our technology and methods to identify women at high risk who could benefit from supplemental imaging.”
Lehman chose to adopt a distinct approach from the conventional computer methods. In the context of individuals with diagnoses, she aimed to shift the focus and ask whether the mammogram could be used not only to detect existing cancer but also for future risk assessment. Behind that premise, Lehman founded Clairity to advance AI-driven healthcare solutions and developed a tool called CLAIRITY BREAST to leverage the power of images to predict the future risk of breast cancer. At the 2025 American Society of Clinical Oncology (ASCO) Meeting, Lehman detailed the announcement that CLAIRITY BREAST has become the first FDA-authorized artificial intelligence (AI) platform for breast cancer prediction, a historic milestone for women’s health.
“As a radiologist, it’s been extremely exciting to spend so much of my career focusing on how to detect disease early, when it can be cured,” said Lehman. “We can now shift our focus even further back, before early detection, to conduct risk assessments as radiologists and identify women who need more advanced screening or will benefit from risk reduction and cancer prevention, which has significantly enhanced our impact.”
The cancer continuum
There is a continuum to the progression of breast cancer, as is true with any disease or cancer; however, different clinical criteria are used to categorize distinct stages. One extreme is a woman whose cancer has progressed to a metastatic Stage IV, and the other is the discovery of breast cancer while it is still in its duct and hasn’t invaded the breast beyond the basement membrane. If breast cancer can be detected at that stage, the chances of survival are incredibly high because numerous therapeutic approaches can kill ductal carcinomas in situ (DCIS) long before they become invasive and spread to other parts of the body.
Lehman illustrates, “I tend to think of cancer as a continuum, with early detection on the left and late-stage disease on the right, where the primary goal is to prolong life. Metastatic diseases are extremely difficult to cure, and it’s all connected. One leads into the other. So, can you move the bar further to the left, even before DCIS, way before it’s an invasive cancer? If you go a little bit further to the left, there isn’t a cancer here yet, but this is the type of patient who is at higher risk of developing breast cancer. We want to approach screening and prevention differently because we don’t screen women at average risk the same way we screen women at high risk. They’re connected, and you’re still doing their risk assessment to identify the best way to screen for early cancer.”
While risk prediction has a slightly prophetic ring to it, that’s essentially what a doctor is doing for other diseases anytime an individual has a checkup. They examine body mass index, blood pressure, cholesterol levels, and other factors to get a rough sense of an individual’s risk for cardiovascular disease, as it is the risk profiles that help guide risk reduction and prevention. Breast cancer has relied primarily on family history and, more recently, genetic screening. An AI platform for image analysis has added a new powerful tool to the toolbox.
“We haven’t been able to help that way very much in breast cancer until now,” said Lehman. “We’re going to do a lot of research, as are others, to evaluate whether this could be a dynamic risk tool, and that really changes things. Could those signals and cues on the mammogram change as a result of interventions, including risk reduction strategies? And would we see that change reflected on the mammogram? That’s gotten us excited.”
Beyond the human eye
The day radiologists switched from developing each image on film to using computers revolutionized the field of medicine. Computers have aided radiologists in speeding up and streamlining imaging, allowing for interaction with the data, including image analysis. Even though AI was first used in radiology to detect tiny calcium deposits in mammograms in 1992, a significant amount of human effort was invested in these methods to define what suspicious areas and lumps looked like on mammograms, as well as to train computers to interpret images like a skilled radiologist would.
However, risk assessment presents a unique challenge. Risk assessment is a domain leveraging current advances in computer vision and artificial intelligence to go far beyond what the human eye can see and use to teach a computer. A simple example of how algorithms “learn” using computer vision is presenting the machine with an immense database of images of cats and dogs. Without being told or taught what a cat or a dog looks like, the computer learns from its neural networks. However, achieving unsupervised learning, where an algorithm can learn without being provided with labels indicating what a dog or cat looks like, has yet to be accomplished in breast cancer.
For CLAIRITY BREAST, Lehman and Clairity had to train an AI algorithm using supervised learning, labeling mammogram images of women years before they developed breast cancer and of women who never developed breast cancer.
“It’s an exciting area for us because it’s allowing AI to go beyond what humans alone have been able to accomplish. We can do a pretty good job of reading mammograms for current cancer. Still, we’ve never been able to do much beyond a very crude breast density measurement to give any sense of whether this woman is at a higher or lower risk. Now we’ve just opened up a new domain to protect against the five-year risk of future breast cancer.”
Despite the relatively simple labeling, CLAIRITY BREAST’s robustness stems from its extensive and diverse image training. By incorporating diversity dimensions such as geography (domestic and international), race, and ethnicity, the algorithm learned how to distinguish what a mammogram looked like in a woman who would develop breast cancer in five years from one in a woman who would not develop breast cancer in that time period.
“I’m particularly thrilled that we were able to include a diverse range of geographic inpatients in our training set and that we set a high standard for demonstrating the effectiveness of this platform,” said Lehman. “Unfortunately, far too often in the past, we’ve left patients out; we just haven’t had the full diversity of patients at risk for disease included in the studies. I’m very happy about that aspect of the inclusivity we had in our patients for this.”
Setting SaMD precedent
Risk models are not typically FDA-regulated. But Lehman and Clairity decided to work closely with the FDA to determine the best path for CLAIRITY BREAST. It was found that there was no appropriate existing product to compare with the breast cancer risk prediction algorithm, so it was classified as software as a medical device (SaMD), specifically a new Class II medical device. If a suitable predicate had existed, Lehman and Clairity could have submitted an application using the 510(k) process. But CLAIRITY BREAST is the first SaMD risk prediction model that generates a score based on the results of a routine mammogram screen. Less than 1% of the FDA’s authorization types are de novo, where there is truly no existing predicate.
“We have predicates for other AI applications in the imaging,” said Lehman. “The first FDA-regulated AI for detection in diagnosis was in 1998. The detection and diagnosis with AI tools have been around for some time. But there is no predicate for this model. It truly opens up a whole new field. We’re excited about this new domain in healthcare of taking an image of any mammogram as input and having as output a risk for predicting the future five-year risk of breast cancer.”
Unlike risk prediction for diseases that have no treatment course, such as Alzheimer’s, currently, the approval of CLAIRITY BREAST could begin to change women’s lives quickly, as this AI platform for identifying future risk of breast cancer plugs right into a well-developed framework of clinical pathways for treating breast cancer. A clinical pathway and infrastructure already exist to assess a woman’s breast cancer risk. Then, the patient’s healthcare provider supports recommendations from various guideline committees regarding how her care should differ from that of women at low risk or average risk. Lehman provided an example in which a woman receives a five-year breast cancer risk score of 3%, indicating a significantly increased risk. Then, the provider would initiate a discussion about the potential benefits of incorporating supplemental testing, such as MRI screening, into the standard routine mammogram screening, alongside the screening mammograms, as part of a long-standing clinical pathway. In addition to early detection methods, there is also a pathway focused on reducing breast cancer risk.
“Breast cancer risk is not a new domain as far as having a risk assessment determine how best to screen for early detection and how best to reduce the risk for cancer prevention,” said Lehman. “What’s new is that we now have a risk model that appears to be identifying women without a family history who are at a higher risk of breast cancer but are not being detected by our traditional risk models. So, I’m excited about the shift from protection and diagnosis to prediction. We have fairly good care pathways; our problem was that we didn’t have risk models. We’ve always wanted to expand our impact beyond early detection to reduce risk and prevent cancer, but we couldn’t. We are excited about these new paradigms and opportunities to explore further the power of imaging, which can detect diseases and predict future outcomes.”
For Lehman and her colleagues, this announcement has been long overdue. Lehman said, “The full team at Clairity, along with our consortium members and friends and colleagues from around the globe who believe in this mission and vision, have contributed significantly to our work and others developing approaches that help us improve methods for reducing risk and screening. This collaboration allows us to support the impact of those advances as well. We’re all working together to find better ways for risk assessment, better ways to screen those at high risk, and better ways to prevent disease from happening in those at high risk.”
