OpenKBP: The open?access knowledge?based planning grand challenge and dataset

The purpose of this work is to advance fair and consistent comparisons dose prediction methods for knowledge-based planning (KBP) in radiation therapy research. We hosted OpenKBP, a 2020 AAPM Grand Challenge, challenged participants develop the best method predicting contoured CT images. models were evaluated according two separate scores: (1) score, which evaluates full 3D distributions, (2) dose-volume histogram (DVH) set DVH metrics. Participants given data 340 patients who treated head-and-neck cancer with therapy. was partitioned into training (n=200), validation (n=40), testing (n=100) datasets. All performed corresponding datasets during phase we ranked based on out-of-sample performance. Challenge attracted 195 from 28 countries, 73 those formed 44 teams phase, received total 1750 submissions. garnered submissions teams. On average, over course improved scores their by factor 2.7 5.7, respectively. In one model achieved significantly better score than runner-up models. Lastly, many top performing reported using generalizable techniques (e.g., ensembles) achieve higher performance competition. This first competition research, it helped launch platform comparing KBP fairly consistently. OpenKBP are available publicly help benchmark future has also democratized research making accessible everyone.