NVIDIA AI Startups Target Breast Cancer Screening and Treatment Gaps

Summary

Breast cancer is the most diagnosed cancer among American women, but most women over 40 skip annual screening. About 40 million U.S. mammograms yearly face a projected shortage of tens of thousands of radiologists over the next decade, while outsourced genomic assays delay treatment by weeks. NVIDIA Inception startup iSono Health’s FDA cleared ATUSA automated wearable 3D ultrasound scans each breast in two minutes versus up to 45 conventionally. Trained on thousands of scans totaling over 1.5 million frames, its GPU accelerated AI standardizes imaging and, CEO Neda Razavi says, is 28% more sensitive than handheld 2D ultrasound. Available in California, Texas, Georgia, Tennessee and Washington, D.C., ATUSA has a 3,200 patient study led by UC Davis and Vanderbilt University Medical Center. iSono’s lesion detection, segmentation and classification AI will expand toward multimodal ultrasound, mammography, MRI and clinical analysis.

Whiterabbit.ai’s FDA cleared WRDensity assesses density and has supported hundreds of thousands; WRRisk estimates long term risk, while new mammography AI seeks more cancer detection and automated negative screening, reducing callbacks, costs and the burden of finding about one cancer per 200 mammograms. Jason Su’s team trains on NVIDIA GPUs at Washington University in St. Louis and the cloud for clinic inference. Ataraxis AI uses existing slides and clinical variables to forecast presurgical chemotherapy response, five year recurrence and postsurgery chemotherapy benefit, where current predictions can require another biopsy and two to four weeks. Joseph Cappadona’s PyTorch and CUDA models, unlike tools trained 15 years ago, are clinically active after validation at more than 10 institutions and multiple trials. SimBioSys models tumors, veins and soft tissue in 3D to guide surgery, estimating recurrence from MRI, pathology and clinical data; CEO Stacey Stevens and NVIDIA’s Chelsea Sumner discussed it October 1 in Phoenix, Arizona. It uses MONAI, cuBLAS, MONAI Deploy and cloud GPUs. Some technologies remain investigational and lack FDA approval.