ESTRO 2024 - Abstract Book

S3879

Physics - Image acquisition and processing

ESTRO 2024

Figure 2 – (Table 1) image quality metrics for the evaluation of the generated images from CycleGANs. Average ± St.Dev. The upper half lists the metrics computed between realistic and original phantom images to assess the anatomical accuracy, the lower half reports the metrics computed on realistic and patient data to evaluate the realistic properties of generated images. a) Average voxels intensity histograms of patient CTs and realistic CTs and b) patient MRIs and realistic MRIs.

MAE_air [H.U.]

MAE_bone [H.U.]

MAE_soft [H.U.]

MAE_tot [H.U.]

RMSE [H.U.] SSIM

PSNR [dB]

NCC

101.61 (9.93)

0.91 (0.02)

65.98 (7.54)

Validation

79.28 (6.10)

112.17 (16.50) 61.86 (6.97)

0.55 (0.06) 26.69 (0.60)

Table 2 - Image similarity metrics between sCT from cGAN and the realistic CT of GT. Average (St.Dev).

Conclusion:

CycleGAN is a feasible approach to generate patient-like data from computational phantoms which can serve as ground truths for the validation of deep learning approaches. The results from sCT generation confirmed a promising generative capability of the net and the feasibility of our approach to validation.

Keywords: validation, synthetic CT, computational phantom

References:

[1] Parrella, G.; Vai, A.; Nakas, A.; Garau, N.; Meschini, G.; Camagni, F.; Molinelli, S.; Barcellini, A.; Pella, A.; Ciocca, M.; et al. Synthetic CT in Carbon Ion Radiotherapy of the Abdominal Site. Bioengineering 2023, 10, 250. https://doi.org/10.3390/bioengineering10020250

1900

Poster Discussion

Anatomy-aware personalization of the clinical target volume for sarcoma patients

Gregory Buti 1 , Chris Beekman 1 , Ali Ajdari 1 , Christopher Bridge 2 , Gregory Sharp 1 , Thomas Bortfeld 1

1 Massachusetts General Hospital, Radiation Oncology, Boston, USA. 2 Massachusetts General Hospital, Radiology, Boston, USA

Purpose/Objective: