ESTRO 2024 - Abstract Book

S4414

Physics - Machine learning models and clinical applications

ESTRO 2024

clinical targets were generated using artificial intelligence based modelling by MVision (MVision, Finland) [8]. Planning target volumes were generated in RayStation using standard clinical margins used in our hospital.

Results:

Validation of the model has resulted in mean dose differences (collapsed cone algorithm) between the sCT and the deformed CT of <0.5% (standard deviation < 0.5%) for the dosimetric parameters D99, D98, D50, D2 and D1.

A comparison of the HU mean average error (MAE) between structures contoured using MVision found that the sCT had a larger MAE for bone type materials (MAE bone = 57HU, MAE femoral heads = 79HU) whilst the MAE for tissue and adipose was comparable (MAE adipose = 6HU, MAE tissue = 13). The dosimetric end to end testing showed mean dose differences of 0.76% (standard deviation 0.77%) between calculated and measured doses. Gamma analysis showed a mean of 99.2% for 3%/2mm between the deformed CT and the sCT.

Conclusion:

Our model produces an sCT with dosimetric results that would be clinically acceptable and help facilitate an MR only pathway for gynaecological patients.

Keywords: synthetic CT, MR-only, dosimetric validation

References:

[1] Jonsson J, Nyholm T, Söderkvist K. The rationale for MR-only treatment planning for external radiotherapy. Clinical and translational radiation oncology. 2019 Sep 1;18:60-5.