Enabling personalized cancer care through AI and computer simulations with NVIDIA

By: Joseph R. Peterson and Dorys Lopez

From Precision Medicine to Personalized Medicine

No two tumors are alike. Despite some tumors sharing the same biological characteristics, inter-patient heterogeneity remains one of the biggest challenges in cancer care. Traditional genomic- and transcriptomic-based precision medicine technologies have made great strides in predicting whether targeted therapies will be effective for a patient. Yet they leave physicians to trial and error in the large populations where rare genetic alterations are absent.

How to better personalize treatment for the general patient population remains an unmet need. What is needed now is a shift from precision medicine to personalized medicine.

Digital Twins as the Basis for Personalized Medicine

SimBioSys® is developing the TumorScope® platform to capture what is truly unique about each patient. The NVIDIA GPU-powered platform takes a holistic and multiscale approach to assess tumor response to therapy. Pulling together all the information that’s painstakingly acquired during the diagnosis — demographic, clinical, radiological and pathological — TumorScope creates a 3D “digital twin” that virtualizes a patient’s cancer.

Once virtualized in a computer, the digital twin can be used to examine what is impossible in the real world: different treatment scenarios on the same patient.

Analogous to the principles of weather forecasting and navigation systems, TumorScope employs mathematical models and computer simulations to forecast tumor response to therapy. Simulations dynamically capture changes in tumor growth, metabolism, drug perfusion kinetics and sensitivity to drugs in both space and time, creating a 4-dimensional (4D) realization of what could happen (Figure 1).



Figure 1. Using diagnostic pre-treatment data, TumorScope creates a 3D digital twin of each patient’s tumor. The digital twin contains dynamic maps of the tumor, blood vessels and surrounding tissue that change in space and time when simulated with different treatment regimens.



Powered by the NVIDIA AI Computing Platform

All of this is made possible by NVIDIA’s full-stack AI acceleration platform. From processing the 3D medical imaging into a realistic model of the patient’s tumor and surrounding tissues, to simulating the complex biophysical processes that drive tumor response, TumorScope was built from the ground up on top of NVIDIA GPUs.

What would have taken days on a supercomputer 10 years ago, can now be accomplished in minutes with NVIDIA GPUs and software development applications.

As a member of NVIDIA Inception, a free program for cutting-edge startups, SimBioSys relied on the expertise of NVIDIA engineers and platforms to drastically optimize code performance. Using the NVIDIA Clara and MONAI frameworks, SimBioSys automated and sped up the time to train new medical imaging networks from about a month to under a day (>30x speedup). By migrating CPU-based simulation code to leverage CUDA, SimBioSys brought down the simulation time for a single patient-specific response prediction from over 10 hours to about 15 minutes (>40x speedup).

These enhancements offer the promise of a non-invasive, point-of-care personalized medicine test.

Taking on Breast Cancer

As a first target, SimBioSys seeks to ensure every breast cancer patient receives the personalized treatment they deserve. Currently, TumorScope supports all early-stage breast cancer standard-of-care therapies listed in the National Comprehensive Cancer Network (NCCN) guidelines.

Utilizing this technology, oncologists will be empowered with information on what treatment could provide the most benefit — and the least amount of toxicity — to better tailor care for each patient (Figure 2).


Figure 2. Upon simulation of the drug regimen chosen by the physician, TumorScope evaluates the changes in size and volume of the tumor to determine a “pCR score,” which reports the probability that a patient’s tumor will be completely eradicated. With this information, oncologists can assess the probability that their treatment of choice could result in pCR for the patient.



TumorScope’s innovative personalized medicine technology has the potential to change the lives of patients for the better. In collaboration with multiple academic and community hospitals, SimBioSys has validated TumorScope in over 1,000 breast cancer patients. The results are state-of-the-art performance in predicting response to therapy: both the size and shape of the tumor throughout treatment and overall response.

SimBioSys is now working to get the technology in physician’s hands so that it can start making a difference in patient’s lives.

Learn the latest about MONAI and innovation from leading healthcare AI startups by attending the panel, Accelerate Healthcare and Life Science Innovation with Makers and Breakers, at NVIDIA GTC, the developer conference for the era of AI and the metaverse.