ESTRO 2023 - Abstract Book

S1333

Digital Posters

ESTRO 2023

geometric agreement with the benchmark structure (DSC<0.6): bowel, cochlea, chiasm, and larynx. There were minimal differences between the tested commercial tools (the mean DSC difference was 0.02). However, scores between software tools were significantly different in 29% of the OARs (bowel, brain, cochlea, eye, lens). Relative dosimetric differences were within 5% for the majority (92%) of OARs and were larger than 10% only for the heart and bowel. Statistical differences in relative dosimetric variations between software tools were observed in 20% of the OARs (bowel, lens, lung, rectum). The mean physician ratings were 3 or higher for all OARs except bowel, bladder, chiasm, and optic nerve (see Figure 1). The largest differences between the AI segmented and manually segmented OARs were due to differences between the clinical protocols used for OAR definition or CT resolution (4 mm slice thickness) in the case of small OARs. A comprehensive analysis of the correlation matrix between the different metrics suggested only a moderate correlation between the geometric metrics and the clinical acceptability (i.e., physician score). Hence, the geometric score should not be taken as a surrogate for clinical usability.

Conclusion A comprehensive study was performed to select and implement commercially available auto-segmentation tools for OAR in all treatment sites in radiation oncology. Several difference metrics were compared, and as a result, the physician score was determined to be the most meaningful metric in determining the performance and usability of the software tools. The investigated AI-based segmentation algorithms need to be improved in the bowel region for clinical use.

PO-1637 Deep learning-based IMRT treatment planning on synthetic-CT for ART in NSCLC-patients

D. Callens 1,2 , L. Vandewinckele 2 , P. Berkovic 1 , F. Maes 3,4 , M. Lambrecht 1,2 , W. Crijns 1,2

1 UZ Leuven, Department of Radiation Oncology, Leuven, Belgium; 2 KU Leuven, Laboratory of Experimental Radiotherapy, Leuven, Belgium; 3 UZ Leuven, Medical Imaging Research Center, Leuven, Belgium; 4 KU Leuven, Processing Speech and Images (ESAT/PSI), Leuven, Belgium Purpose or Objective Patients with NSCLC-tumours could benefit from adaptive radiotherapy. To increase the efficiency of an adaptive workflow, daily CBCT-imaging could be used for treatment planning. Our group reported on CT-based treatment planning by autoplanning for lung IMRT using DL-based fluence map generation. This work investigates the performance of this treatment plan prediction on acquired CBCT images. The CBCT images are converted to syntheticCT’s (sCT) utilising two separate commercial solutions. The generated treatment plans are compared with semi-automatic planning created on the sCT’s. Materials and Methods For this retrospective study, 9 NSCLC-cases were selected out of a recent database of stage III/IV-patients. These cases were chosen upon fractionation scheme (60/66Gy in 30/33Fx). The mid-treatment CBCT was non-rigidly registered with planningCT to create sCT’s. Deformations were done in MIM and Velocity. Target structures of planningCT were deformed accordingly. Syngo.via was used to autocontour OAR. Three independently fluence-prediction networks with different inputs (deformed structures [RS]; sCT; initial beam dose, rigidly transformed [D]) delivered fluences used for dose computations of the DL-based plans in Eclipse TPS. Semi-automatic planning through RapidPlan mimics standard-of-care planning workflow. All plans started from an in-house IMRT class-solution with 9 beams. Plan- and DVH-parameters were used for plan comparison. Results Both sCT’s could be used to generate structures necessary for the clinical standard planning as well as the fluence prediction. The current planning workflow outperformed fluence prediction meeting all clinical constraints. In this case, dose maximums, HI- and CI-values were similar on both sCT’s. In total, 12 of 54 DL-plans met all DVH-constraints. DL-based planning often resulted in an inferior coverage of PTV and a higher dose on OAR. Inferior coverages occurred mainly when visible tumor deformation was observed. MIM noticeably deformed more than Velocity, although both transformation models are accompanied with uncertainty in truthfulness of deformation. Input for the networks is of influence. The combination D+RS led to plans with higher target coverage together with lower doses on OAR and equally acceptable HI/CI, in comparison with other inputs. The combination sCT+D+RS as input resulted repeatedly in unacceptable plans due to a high hotspot outside targets.