Can knowledge-based DVH predictions be used for automated, individualized quality assurance of radiotherapy treatment plans?

Table 1 Summary of the RapidPlan model characteristics for each individual organ-at-risk (OAR)

Structure name	Number of structures included in model	Suggested number in model	Model fit (R²)	Regression model’s average χ²
Contralateral Parotid	90	-	0.77	1.096
Contralateral Submandibular	69	-	0.68	1.094
Ipsilateral Parotid	86	-	0.52	1.051
Cricopharyngeal Muscle	52	-	0.75	1.117
Lower Larynx	46	-	0.82	1.125
Upper Larynx	26	36	0.76	1.196
Inferior PCM^a	38	48	0.86	1.229
Medial PCM^a	24	42	0.58	1.169
Superior PCM^a	39	48	0.89	1.128
Upper Esophageal Spincter	68	-	0.85	1.147
Oral Cavity	76	-	0.80	1.073

Abbreviation: ^a PCM Pharyngeal constrictor muscle
The model fit R² value shows the correlation between the geometric and dosimetric regression parameters. Where relevant, the RapidPlan model configuration algorithm can suggest to include more OARs in the model library. This could for example be in the case where there is large variation in OAR sparing between the plans that are included in the model library. Including more plans could improve the prediction accuracy. Because not every structure was attempted to be spared in the original clinical plan, the number of structures included in the model varies per OAR

ISSN: 1748-717X