Science

Can We Trust an AI Oncologist to Help Cancer Patients?

tumor

Before radiation therapy, oncologists have to carefully review medical images of the patient in order to identify the gross tumor volume – that is, the portion of the disease that can be observed. Once they have that data, they can design patient-specific radiation therapy.

The process is widely known as contouring and it establishes how much radiation a patient needs to receive and the way it will be delivered. The neck and the head are very vulnerable to this particular treatment due to the delicate tissues in the vicinity.

Unfortunately, contouring is a subjective issue, meaning some doctors can suggest risky target volumes for the patient. It goes without saying that his variability is a problem for the patients, who might be over or under-dosed.

Recently, Carlos Cardenas, a graduate research assistant and PhD candidate at The University of Texas MD Anderson Cancer Center in Houston, Texas, alongside a team of researchers, managed to develop a new method to automate the contouring.

How? With artificial intelligence and deep neural networks.

Their work focuses on translating the process into a computer program, based on the clinical data and radiation therapy treatment plan data at MD Anderson. All those things put together can replicate physician patterns used to treat certain types of tumors.

Cardenas has gathered data since 2015, when he began the project, and added it all to the deep learning algorithm he developed, which in turn helped him identify and recreate physician contouring patterns.

Him and his collaborators tested the AI on cases that had been left out of the training data and soon found out that the AI’s results were comparable to those of trained oncologists. The contours predicted were in agreement with the human decisions and had every chance to be implemented with almost no changes.

The method stands out not only through its brilliance but also due to its speed and efficiency – it takes the Maverick supercomputer at the Texas Advanced Computing Center just under a minute to produce clinical target volumes.

Cardenas would like to use the project to help low-and-middle income countries that do not have easy access to expertise contouring.

If you are curious to know more, their results are reported in the June 2018 issue of the International Journal of Radiation Oncology *Biology* Physics.

Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

To Top