ESTRO 2024 - Abstract Book

S3832

Physics - Image acquisition and processing

ESTRO 2024

Complete results are given in Table 1 alongside an indication of significant differences.

MAE significantly differed between all algorithms, with syngo.via VB60 yielding the smallest error (55 HU; Table 1). Detailed evaluations showed that MriPlanner performed the best in terms of MAE and DSC for bony structures. For these same metrics, syngo.via VB40 performed significantly worse than the two deep learning-based algorithms. Mean gamma passing rates for the high-dose region (D>90%) were equal to 100% for all algorithms. At a threshold of D>10%, mean passing rates ranged from 96.3% to 97.4 %, respectively for syngo.via VB60 and VB40, with the former performing significantly worse than the other algorithms (Table 1). Target DVH parameters were slightly, but consistently, overestimated for all three algorithms compared to CT. Discrepancies were more pronounced for the highest doses, with the maximum deviation found for D2% of the boost CTV (0.6Gy = 0.8%, Table 1). Conversely, no significant differences were found for D98% within the Planning Target Volume (PTV). The remaining comparisons for target volumes showed syngo.via VB40 significantly differing from CT. MriPlanner could not be shown to significantly differ from CT or syngo.via VB60 in any comparison. Neither were any significant differences observed in the organ at risk parameters.

Conclusion:

While some discrepancies in image fidelity and dosimetric validity could be demonstrated, their magnitudes were generally low which resulted in a limited clinical significance and proved the overall equivalence of the investigated software.