Author: Lisa Kuhns
Machine learning models achieve clinically acceptable accuracy in image segmentation tasks in radiotherapy planning and reduce overall contouring time for head and neck and prostate cancers, according to a recent study in JAMA Network Open (2020;3:e2027426. doi:10.1001/jamanetworkopen.2020.27426)
Personalized radiotherapy planning requires large time commitments for oncologists and processes often vary among experts and institutions.
Authors aimed to explore clinically acceptable autocontouring solutions that can be integrated into clinical practice and used in different radiotherapy areas.
Researchers evaluated multicenter imaging data set made up of 519 pelvic and 242 head and neck computer tomography scans from 8 clinical sites. Patients in the study were diagnosed with either prostate or head and neck cancer. The models were trained to automatically delineate organs at risk and evaluated internal and external datasets. Models were compared against expert annotations in an interobserver variability (IOV) study.
For 13 of the 15 structures, the models performed within the bounds of expert IOV. For internal vs external data sets, the models achieved mean [SD] Dice scores for left femur at 98.52% and 98.04% (P = .04), respectively.
“In this study, the models achieved levels of clinical accuracy within expert IOV while reducing manual contouring time and performing consistently well across previously unseen heterogeneous data sets,” concluded the study authors. “With the availability of open-source libraries and reliable performance, this creates significant opportunities for the transformation of radiation treatment planning.”—Lisa Kuhns