NVIDIA reported that startups in its Inception program are developing artificial intelligence tools to address clinical bottlenecks across breast cancer screening, imaging, and treatment planning. About 40 million mammograms take place in the United States every year, facing a projected shortfall of tens of thousands of radiologists over the next decade. At the same time, traditional genomic assays sent to external laboratories require weeks to process.
iSono Health developed the FDA-cleared ATUSA platform, a wearable 3D ultrasound system designed to simplify imaging. The device captures a standardized breast volume in approximately two minutes per breast, compared with up to 45 minutes for conventional handheld ultrasounds. The system's AI was trained on thousands of full-breast scans comprising more than 1.5 million ultrasound frames. Using NVIDIA graphics processing units, the company reports its 3D scan is 28 percent more sensitive than handheld 2D ultrasound. ATUSA is commercially available in clinics across California, Texas, Georgia, Tennessee, and Washington, D.C. The startup is currently conducting a 3,200-patient multicenter clinical study with lead research sites at UC Davis and Vanderbilt University Medical Center.
Automated screening
Whiterabbit.ai built FDA-cleared software called WRDensity, which automatically assesses breast density from mammograms and has been used in care for hundreds of thousands of patients. The company also developed WRRisk to estimate long-term breast cancer risk, while researching models to identify negative scans and flag tumors. Whiterabbit trains its models on an NVIDIA GPU cluster at Washington University in St. Louis and runs clinical inference directly on local hardware. Jason Su, cofounder and chief technology officer of Whiterabbit.ai, noted that breast radiologists must locate roughly one cancer in every 200 mammograms.
Treatment predictions
Ataraxis AI is building software to evaluate digital pathology slides and standard clinical data, aiming to bypass separate tissue biopsies that carry two- to four-week waiting periods. One of its models predicts whether presurgical chemotherapy will shrink tumors, while another estimates five-year recurrence risk and post-surgical chemotherapy benefit. Both models run on NVIDIA GPUs using PyTorch and CUDA, and have undergone validation across more than 10 institutions and multiple clinical trials.
SimBioSys produces 3D models of breast tumors, veins, and soft tissue using MRI, pathology, and clinical data to assess recurrence risk. The company trains its models with NVIDIA MONAI and CUDA-X libraries in the cloud. Stacey Stevens, chief executive of SimBioSys, stated that the computing architecture allows clinicians to analyze hundreds or thousands of images quickly rather than waiting weeks for answers. NVIDIA noted that certain technologies detailed across these startup programs remain investigational and have not received FDA approval for commercial use.
