Sign In

Communications of the ACM

ACM TechNews

AI Oncologist to Help Cancer Patients Worldwide

View as: Print Mobile App Share:
Comparison between computer-predicted ground-truth clinical target volume (blue) and physician manual contours (red) for three oropharyngeal cancer patients.

University of Texas researchers have devised a way to automate the contouring of tumor volumes.

Credit: Texas Advanced Computing Center

University of Texas MD Anderson Cancer Center researchers devised a new way to automate the contouring of tumor volumes using artificial intelligence and deep neural networks.

Carlos Cardenas and his colleagues analyzed data from 52 oropharyngeal cancer patients whose gross tumor volumes and clinical tumor volumes had been contoured for radiation therapy treatment.

By observing the radiation oncology team at MD Anderson, Cardenas determined how they define targets, then turned to deep learning to reveal the unwritten rules guiding the experts' decisions.

Cardenas wrote an algorithm that uses auto-encoders, neural networks that can learn how to represent datasets, to identify and recreate physician contouring patterns.

Tests show the model's results are comparable to the work of trained oncologists, which could lower inter-physician variability by increasing standardization in contouring, allow comparisons of outcomes in clinical trials, and significantly boost efficiency.

The team obtained clinical target volumes in less than a minute using the Maverick supercomputer at the Texas Advanced Computing Center.

From Texas Advanced Computing Center
View Full Article


Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA


No entries found