Deep learning model for automatic contouring of cardiovascular substructures on radiotherapy planning CT images: Dosimetric validation and reader study based clinical acceptability testing

• Published in: Radiotherapy and oncology Vol. 165; pp. 52 - 59
• Main Authors: Garrett Fernandes, Miguel, Bussink, Johan, Stam, Barbara, Wijsman, Robin, Schinagl, Dominic A.X., Monshouwer, René, Teuwen, Jonas
• Format: Journal Article
• Language: English
• Published: Ireland Elsevier B.V 01.12.2021
• Subjects: Automatic cardiac contouring; Carcinoma, Non-Small-Cell Lung; Cardiac radiotoxicity; Deep Learning; Hematology, Oncology, and Palliative Medicine; Humans; Lung cancer; Lung Neoplasms; NSCLC; Organs at Risk; Radiotherapy Planning, Computer-Assisted; Segmentation; Surface dice; Tomography, X-Ray Computed; LL; OAR; SVC; RT; RV; Automatic cardiac contouring; Segmentation; Lung cancer; DL; GTV; LV; NCTN; HT; Cardiac radiotoxicity; NTCP; CA; NSCLC; OS; Surface dice; AO; RA; MAE; CT; PA; RC; LA; NCI; DVH; IVC; RL; pulmonary artery; left ventricle; normal tissue complication probability; national cancer institute; lung cancer; gross tumor volume; Cardiac Radiotoxicity; overall survival; Automatic Cardiac Contouring; dose-volume histogram; non-small cell lung cancer; left lung; Deep Learning; heart; requires corrections; Tumor Adjacency; Residual U-net; radiotherapy; right lung; Surface Dice; left atrium; National Clinical Trials Network; clinically acceptable; computed tomography; mean absolute error; inferior vena cava; right ventricle; right atrium; superior vena cava; aorta; organ at risk

•Accurate automatic contouring is essential for cardiac radiotoxicity research.•Developed residual U-net model achieves high geometric and dosimetric accuracy.•Model validated in multi-institutional NSCLC dataset.•Model robust against adjacency of tumors to contoured structures.•Average clinical acc...