ESTRO 2024 - Abstract Book

S3615

Physics - Dose prediction, optimisation and applications of photon and electron planning

ESTRO 2024

Figure 1A shows the ipsilateral lung mean predicted dose of Consortium A KB models on the Consortium B data set, with SDmean=1.1Gy to be compared with a standard deviation of the clinical ipsilateral lung mean predicted dose (averaged on the Consortium B data set) of cSDmean=0.6Gy. With regards to the ipsilateral lung V20Gy prediction, figure 1B shows the results with an average SDV20Gy=2.7%. The clinical Consortium B external cross validation data set ipsilateral lung V20Gy shows a cSDV20Gy=1.6%.

Figure-1: boxplot of mean predicted ipsilateral lung mean dose and V20Gy of Consortium B external cross validation patient data set on the Consortium A models. Red dotted lines represent the average mean dose (A) and the average V20Gy (B) of the clinical Consortium B patient data set values. Figure 2 (left) shows the fraction of Consortium A and B cross-validation patients test with PC1 within the 90th percentile of the training set for all Consortium A KB models. Intra-institutional validation of Consortium A reported high inter-institute interchangeability, with an ipsilateral lung PC1 value inside the 90th in 9 out of 10 models. The value of PC1 for each model of Consortium A within the international cross-validation patients’ cohort of Consortium B, was inside the 90th percentile in 6 out of 10 models. PC1 for Institute 6 showed poor transferability both for the intra-institutional validation of Consortium A (7 out of 18 patients) and international inter-institutional validation of Consortium B (7 out of 20 patients), which also showed poor transferability for Institute 10 (5 out of 20 patients). With regards to the 20 external cross-validation patients test of Consortium B, the ipsilateral lung DVHs prediction performance of Consortium A KB models was quantified (figure 2 right) resulting in 76% of optimal prediction, 16% of improved prediction, 7% in suboptimal prediction and 1% of failed prediction.