How AI Is Helping Improve Breast Cancer Screening, Diagnosis and Treatment

Breast cancer is the most commonly diagnosed cancer among American women, yet major disparities remain in screening and treatment. The majority of women over 40 skip recommended annual mammograms. At the same time, radiologists are reading more mammograms with fewer colleagues, and it can take several weeks for diagnostic tests and results to inform treatment decisions.

Companies in the NVIDIA Inception startup program are developing AI applications to support clinicians at each of these points, including medical imaging, breast cancer risk assessment and treatment planning.

Approximately 40 million mammograms are performed in the United States each year. However, a shortage of tens of thousands of radiologists is predicted over the next decade, placing additional strain on the healthcare system’s ability to read mammograms.

At the other end of the treatment timeline, decisions often depend on genomic assays sent to outside laboratories. These assays can take several weeks to process, delaying answers at a time when speed and certainty are especially important.

The following NVIDIA Inception companies are working to address challenges across the breast cancer care journey, from screening and imaging to diagnosis, treatment planning and recurrence-risk assessment.

AI-powered imaging brings breast ultrasound closer to patients

Access is a significant barrier for women who have difficulty finding time or convenient locations for breast cancer screening. Missed screenings can lead to delayed diagnoses. NVIDIA Inception startup iSono Health was founded to simplify the breast imaging workflow.

The company’s FDA-cleared ATUSA platform is a wearable, automated 3D quantitative ultrasound system. It captures standardized breast volumes in approximately two minutes per breast. By comparison, traditional handheld ultrasound examinations can take up to 45 minutes.

ATUSA’s AI was trained on thousands of whole-breast scans comprising more than 1.5 million ultrasound frames. The system uses NVIDIA GPU acceleration and open-source medical imaging technology to automate image acquisition and produce 3D scans that are 28% more sensitive than handheld 2D ultrasound.

Because handheld ultrasound depends on how a person holds and moves the probe, it is generally difficult to compare a woman’s scans from one year to the next. ATUSA captures the entire breast in a consistent way, creating a reproducible view of breast tissue. This can help clinicians assess how tissue changes over time while reducing operator error and variability.

ATUSA is sold through partner clinics in California, Texas, Georgia, Tennessee and Washington, D.C., with new locations coming online regularly.

“Bringing scans closer to patients is the first breakthrough,” said Neda Razavi, CEO of iSono Health. “Our vision is to make scans increasingly informative, allowing clinicians to see what is there, understand what has changed, and make more informed decisions.”

iSono Health has developed AI capabilities for lesion detection, 3D segmentation and lesion classification. The company plans to extend its AI pipeline to multimodal diagnostic intelligence across 3D ultrasound, mammography, MRI and clinical information.

To further validate the platform’s performance, iSono Health is conducting a multicenter clinical study of 3,200 patients. Study centers include the University of California, Davis, and Vanderbilt University Medical Center.

AI helps detect breast cancer while reducing false alarms

Whiterabbit.ai, another NVIDIA Inception company, is developing AI technology for breast cancer screening. Its FDA-cleared WRDensity software automatically assesses breast density from mammograms and is used in the care of hundreds of thousands of patients.

The company is also developing WR Risk, clinical decision-support software designed to estimate a patient’s long-term risk of developing breast cancer. In addition, Whiterabbit.ai is researching a new generation of mammography AI intended to help radiologists detect more cancers while automating the screening of mammograms that appear negative.

The goal is to ease pressure on the radiologist workforce, speed up results, reduce avoidable patient callbacks and lower downstream healthcare costs.

“Every day, breast radiologists face a needle-in-a-haystack problem trying to find approximately one cancer in every 200 mammograms,” said Jason Hsu, co-founder and chief technology officer of Whiterabbit.ai. “We hope that AI will become a powerful companion for radiologists, helping them cut through the hay and allowing them to focus their expertise on the areas that matter most.”

Whiterabbit.ai trains its AI models on a cluster of NVIDIA GPUs at Washington University in St. Louis, supplemented by additional GPU capacity in the cloud. Inference runs on NVIDIA GPUs deployed directly in clinics.

Predicting which breast cancer treatments may work

After a breast cancer diagnosis, one of the most important questions is how the cancer will respond to treatment. Answering that question often requires additional testing, and the range and accuracy of current predictions remain limited.

Another tissue biopsy may be required, and the results can take two to four weeks. Ataraxis AI uses digital data, including pathology slides that are already part of a standard patient workup, to develop clinical intelligence that predicts patient outcomes and responses to different treatments.

“Most of the tools that oncologists rely on today to make treatment decisions were only trained once 15 years ago and never updated,” said Joseph Cappadona, technical staff member at Ataraxis AI. “As we acquire more clinical trial data, our models are enhanced. But as we expand the model, the bigger change is that we will be able to answer more questions to allow oncologists to individualize treatment across all cancers.”

Ataraxis AI’s models analyze digital pathology slides and standard clinical variables to predict treatment response and recurrence risk. One model predicts whether neoadjuvant chemotherapy is likely to shrink a patient’s tumor before surgery. After surgery, another model estimates a patient’s five-year recurrence risk and the potential benefit of chemotherapy as a next step.

Both models have been validated in more than 10 institutions and multiple clinical trials and are actively used in clinical practice. They run on NVIDIA GPUs on-premises, in off-site data centers and in the cloud using PyTorch accelerated by NVIDIA CUDA.

3D visualization offers new insight into breast tumors

Another NVIDIA Inception company, SimBioSys, joined NVIDIA for a panel discussion commemorating Breast Cancer Awareness Month.

SimBioSys develops AI-powered precision medical technology that creates 3D models of breast tumors, veins and other soft tissues. These models provide insights that can influence surgical guidance and treatment planning. The company has also developed a tool to estimate breast cancer recurrence risk using 3D volumetric data from breast MRI, tumor pathology and clinical information.

“We are currently building a platform that can take multimodal data, including imaging tests, pathology results, genomic tests where applicable, and other biological inputs, and integrate it all using AI,” said Stacey Stevens, CEO of SimBioSys, at the event. “Doing so creates new insights beyond what you get from individual pieces.”

SimBioSys uses NVIDIA MONAI for training and validation data. The company also uses NVIDIA CUDA-X, which includes libraries such as cuBLAS, and MONAI Deploy. Its imaging technologies running on NVIDIA GPUs in the cloud will be evaluated.

“NVIDIA technology gives us the computing power to take hundreds or thousands of images and apply AI to quickly analyze them,” Stevens said. “This speed is important because patients and doctors need answers right away. They can’t afford to wait days or even weeks.”

For more information, visit the NVIDIA Inception program for startups.

The specific technology described in this article is under investigation and has not been approved for commercial use by the U.S. Food and Drug Administration.