Automated Organ Segmentation with Deep Learning Algorithm

Automated Organ Segmentation with Deep Learning Algorithm

Automated Organ Segmentation with Deep Learning Algorithm: A Pilot Study of Artificial Intelligence in the Radiation Treatment Planning Process for Esophageal Cancer.

Dr. Ming Pan

Radiation Oncologist,
Windsor Regional Hospital

Arash Ahmadi

Electrical and Computer Engineering,
University of Windsor

John Agapito

Medical Physicist,
Windsor Regional Hospital

FUNDER: Windsor Cancer Centre Foundation, Seeds4Hope Program


Related Programs:
Nucleus Cores:

Esophageal cancer is an aggressive malignancy. In 2017, an estimated 2,300 Canadians were diagnosed with it and 2,200 Canadians died from it. External beam radiation treatment (EBRT) has become standard for these patients, including preoperative, postoperative, radical treatment for inoperable cases, and palliative treatment for symptom control.  Windsor Regional Hospital provides all the EBRT for our local/regional patients so that they can receive excellent care close to home. But often they have to wait a long time before starting treatment. This is because we need to contour all structures on Computerized Tomography (CT) scans before asking the computer to design beams and find a solution to deliver high dose to the cancer without damaging the other important organs. This is a laborious process that often delays plan design and analysis, and prolongs patient’s wait time.

Our goal is to develop an Artificial Intelligence (AI) computer to eventually automatically contour organs and structures on our treatment planning CT scans, thus potentially improve patient care and shorten wait times.

We will use CT datasets to build an image library for training AI for hundreds of times. The algorithm will be tested with new images and receive feedback for further training. We will then evaluate the AI-generated structures by comparing to those contoured by our expert team members including an experienced radiation oncologist, a physicist, an engineer, and a computer scientist.

The result of this study can support ongoing research to improve AI for Quality Assurance (QA) purposes. In the future it might provide feedback to the treatment planning team to modify the original contouring of the structures of interest, thus improve esophageal cancer treatment and outcome. This might take years to happen, but in the meantime our pilot study might help shorten the wait time to start EBRT in this community.