Radiotherapy is a well established method for attacking tumors within the body. There are a number of techniques that are used to administer radiation to a lesion, but they all come with the risk of injuring nearby tissues and organs. Gamma beams and other directed high energy devices result in the exposure of all the tissues that are on the way to and on the other side of a target, which is a serious problem. Knowing where the important organs are in individual patients can allow clinicians to prepare radiation therapy treatments so that as little collateral damage occurs as possible. CT scans are usually used to map the internal anatomy. Currently, this is a job for radiation oncologists and it is rather difficult to properly trace the relevant organs and provide a treatment path that will minimize damage.
A team of researchers from University of California Irvine, Shanghai Jiao Tong University School of Medicine in China, and DeepVoxel, a company based in Costa Mesa, California, has now unveiled a system that takes a CT scan as an input and automatically provides an outline of all the important organs within it. The technology should help to speed up radiotherapy planning, improve clinical confidence, and hopefully result in less damage to internal organs.
The new system relies on deep-learning methods to process a scan within seconds, something that normally takes over a half hour to perform manually. “On a data set of 100 CT scans, our deep-learning method achieved an average similarity coefficient of more than 78 percent, a significant improvement over analyses done by radiation oncologists,” said Xiaohui Xie, one of the co-authors of the study appearing in journal Nature Machine Intelligence.
The system works with data from a variety of manufacturers, including low contrast CT, and doesn’t require particularly powerful computers to do its job. As such, it should be easy to roll it out to radiation therapy facilities and get clinicians acquainted with using it.
Study in Nature Machine Intelligence: Clinically applicable deep learning framework for organs at risk delineation in CT images