Artificial intelligence in breast imaging is rapidly expanding beyond computer-aided cancer detection.

Today’s algorithms can characterize mammographic findings, quantify breast density, assist with workflow triage and technical quality assurance, and extract imaging features associated with future breast cancer risk.

The broader goal is increasingly ambitious: moving from predominantly age-based screening toward a more individualized model that integrates imaging phenotype, clinical risk and, potentially, genetic information.

For referring physicians, that evolution could influence how patients are stratified for surveillance and supplemental imaging.

Kathy J. Schilling, M.D., FACR, diagnostic radiologist and medical director of the Christine E. Lynn Women’s Health & Wellness Institute at Boca Raton Regional Hospital, part of Baptist Health, has been involved in evaluating and implementing emerging breast-imaging technologies.

“Breast radiologists first started talking about the use of artificial intelligence in our practices 10 years ago. Since then, we have seen its implementation in many aspects of our practice at Lynn Women’s Institute,” Dr. Schilling says.

From Lesion Detection to Imaging Phenotype

Cancer detection and characterization remain among AI’s most established applications in mammography. Current systems can analyze an examination, identify individual findings and assign levels of suspicion to both the study and specific abnormalities.

But AI can also extract information beyond visible lesions.

Algorithms can perform volumetric assessment of fibroglandular tissue, providing an objective measure of breast density at each examination. More sophisticated systems can interrogate features difficult for the human observer to quantify reliably, including:

• distribution of fibroglandular tissue;
• parenchymal patterns;
• calcifications and masses;
• asymmetry; and
• overall mammographic complexity and texture.

“These findings are identified by AI but defy identification by the human eye and brain,” Dr. Schilling says.

Together, these features can contribute to a mammographic risk phenotype — an imaging-derived estimate of a woman’s likelihood of developing breast cancer despite a currently normal mammogram.

Mammography as a Dynamic Risk Biomarker

Traditional breast cancer risk models incorporate clinical variables such as age, reproductive history and family history, with breast density included in some models. AI-based mammographic risk assessment approaches the problem differently.

Following a normal mammogram, an algorithm can compare a patient’s imaging phenotype with large databases of mammograms from women who subsequently developed breast cancer. Similarity to higher-risk patterns can then be used to estimate short-term risk over the next several years.

Importantly, mammographic risk is not static. AI-derived risk may change between examinations, potentially adding longitudinal information that conventional risk models do not capture as readily.

That raises the possibility of mammography serving two functions simultaneously: detecting an existing malignancy and providing a dynamic biomarker of future risk.

Toward Risk-Based Screening

Research suggests that combining mammographic risk with clinical information improves predictive performance, and Dr. Schilling says incorporating genetic information may eventually refine stratification further.

“AI is giving the breast radiologist the ability to move from traditional age-based screening for every woman to risk-based screening from information gained from clinical findings, genetics and information embedded in the mammogram,” she says.

The potential implication is more individualized use of supplemental imaging.

High breast density is a recognized breast cancer risk factor and decreases mammographic sensitivity, but density alone does not identify every woman at elevated risk. An imaging-derived model could potentially identify higher-risk women with nondense breasts while also helping determine which women with dense breasts are most likely to benefit from MRI, ultrasound or intensified surveillance.

Conversely, patients consistently classified as lower risk might eventually be candidates for less intensive screening intervals.

Such approaches remain under investigation. Before screening frequency or modality is changed on the basis of AI-derived risk, clinical studies will need to demonstrate that improved risk prediction translates into better patient outcomes.

Workflow, Quality and Resource Allocation

AI is also affecting breast imaging operations.

Algorithms can assign a case score reflecting the likelihood of suspicious findings, potentially allowing low-suspicion examinations to be reviewed more efficiently while directing greater physician attention to complex or higher-risk studies.

Other applications include structured report generation and automated quality assessment. AI can evaluate image positioning, breast compression and other technical parameters, helping practices monitor performance and maintain Mammography Quality Standards Act requirements.

Algorithms can also identify breast arterial calcifications, potentially providing opportunistic information associated with cardiovascular risk.

More broadly, effective risk stratification could help imaging programs direct limited resources toward patients most likely to benefit from additional surveillance.

What AI Is Changing in Practice

Dr. Schilling says her practice has already observed benefits from AI-supported interpretation.

“We have witnessed the impact AI has had on our cancer detection rate, especially on invasive cancers and interval cancers — those diagnosed between scheduled screening examinations,” she says.

She also reports reductions in false-negative and false-positive examinations, fewer recalls and biopsies, and improvements in radiologist efficiency and confidence.

For physicians, the relevant measure of AI’s value is not simply whether an algorithm identifies more abnormalities. It is whether the technology helps detect clinically significant cancers earlier, reduces unnecessary diagnostic workups and enables radiologists to focus their attention where it is most needed.

The Physician Remains Central

The performance of breast imaging AI remains dependent on the quality and diversity of the data used to develop and validate individual algorithms.

“It is important to know the AI can continually be improved with exposure to additional high-quality data sets,” Dr. Schilling says.

Before AI-derived risk routinely guides screening intervals or supplemental imaging, algorithms will require continued validation across women of different ages, races, breast densities and underlying risk profiles.

The goal is not to replace physician judgment. Rather, AI may provide an additional layer of information to be considered alongside personal and family history, breast density, genetic information, prior imaging and patient preferences.

For breast imaging specialists and referring physicians, that may ultimately be AI’s most significant contribution: expanding mammography from a test focused primarily on whether cancer is visible today into one that may also help identify which patients warrant greater vigilance tomorrow.