NVIDIA spotlights startup AI aimed at closing critical bottlenecks in breast cancer care

Breast cancer is the most commonly diagnosed cancer among American women, yet the care pathway can be slow and uneven, with delays that may affect outcomes. In a new post focused on gaps from first scan to treatment planning, NVIDIA highlights how several startups supported through NVIDIA Inception are applying artificial intelligence to reduce friction points at both ends of the timeline, including screening and genomic or tissue-based testing that can take weeks to complete.

NVIDIA’s overview points to participation and capacity constraints in screening. It says a majority of women over age 40 skip the recommended annual screening, while radiologists are interpreting increasing numbers of mammograms with fewer colleagues. When a diagnosis does arrive, the next steps can stall: tests used to inform treatment can take weeks to return results, according to the company.

The companies NVIDIA features are attempting to address these delays and uncertainty with AI tools designed to be faster, more consistent, and more directly connected to clinical decision-making. “These startups are building AI applications to support clinicians at each of these friction points, including imaging, risk assessment and treatment planning,” NVIDIA wrote, describing the work as accelerated by NVIDIA AI infrastructure.

One example is iSono Health, an NVIDIA Inception startup focused on bringing imaging closer to patients. The company built its FDA-cleared ATUSA platform, described by NVIDIA as a wearable, automated 3D quantitative ultrasound system. NVIDIA says ATUSA captures a standardized breast volume in approximately two minutes per breast, compared with up to 45 minutes for conventional handheld ultrasound.

NVIDIA attributes much of the system’s performance to automation and compute. The blog says ATUSA’s AI, trained on thousands of full-breast scans comprising more than 1.5 million ultrasound frames, automates image acquisition using NVIDIA GPU acceleration and open-source medical imaging technology. NVIDIA also says the platform is “28% more sensitive” than handheld 2D ultrasound and is intended to produce a repeatable view of breast tissue that can help clinicians analyze changes across successive scans, potentially reducing operator variability.

In addition to deployment through partner clinics, NVIDIA describes how iSono is extending its workflow from scanning to interpretation. It says iSono has developed AI capabilities for lesion detection, 3D segmentation and lesion classification, and that it plans to expand into multimodal diagnostic intelligence covering 3D ultrasound, mammography, MRI and clinical information. NVIDIA also reports a multicenter clinical study with 3,200 patients underway, with lead sites at UC Davis and Vanderbilt University Medical Center.

Other NVIDIA Inception companies highlighted focus on screening workflow and the problem of limited detection capacity. ’s FDA-cleared WRDensity software automatically assesses breast density from mammograms, NVIDIA says, and has been used in the care of hundreds of thousands of patients. NVIDIA adds that Whiterabbit also developed WRRisk, a clinical decision support tool estimating long-term risk of developing breast cancer, and it is researching a next-generation AI approach for mammography intended to detect more cancers and automate evaluation of negative screens.

NVIDIA frames the screening challenge as a “needle-in-a-haystack” task for radiologists and quotes Whiterabbit’s cofounder and chief technology officer, Jason Su, describing cancer detection occurring in roughly one case per 200 mammograms. NVIDIA says Whiterabbit trains its models on a cluster of NVIDIA GPUs at Washington University in St. Louis, with additional GPU capacity in the cloud, and runs inference on NVIDIA GPUs deployed in clinic settings.

On the treatment side, NVIDIA points to Ataraxis AI as addressing delays and uncertainty in prediction once cancer is diagnosed. It says treatment decisions often depend on tissue biopsy and related tests, with a two- to four-week wait, and that Ataraxis is building clinical intelligence to predict patient outcomes and response to therapies using digital data, including pathology slides already part of standard workups. NVIDIA describes models that estimate whether presurgical chemotherapy is likely to shrink a tumor enough to indicate a response before surgery, and models that estimate five-year recurrence risk and likely benefit from chemotherapy after surgery. NVIDIA says these models have been validated across more than 10 institutions and multiple clinical trials, and are in active clinical use, running on NVIDIA GPUs in settings including on premises, an offsite data center and the cloud, with PyTorch accelerated by NVIDIA CUDA.