ESTRO 2024 - Abstract Book

S3756

Physics - Image acquisition and processing

ESTRO 2024

the center data from the two middle quartiles of age, while the remaining data was reserved for evaluating image quality and generalization. In all experiments, a standard 80/20 train and test split was performed.

Results:

The Mean Average Error (MAE), Peak Signal to Noise Ratio (PSNR), Root Mean Square Error (RMSE), and Structural Similarity Index Measure (SSIM) metrics are employed for the quantification of image quality improvement resulting from the proposed network. These metrics were computed exclusively on the patient's body.

Evaluation of both sexes using the model trained on men reveals similar results across the selected metrics, along with their respective standard deviations. Specifically, the metrics MAE, PSNR, RMSE, and SSIM values for men were 23.62 ± 8.88 HU, 22.92 ± 1.98 dB, 76.02 ± 18.65 HU, 0.852 ± 0.042. For women, the corresponding values were 24.19 ± 18.24 HU, 22.94 ± 3.16 dB, 78.60 ± 35.21 HU, 0.864 ± 0.062. Furthermore, the qualitative analysis suggests that the generated skVCT images exhibit similarity for both sexes and represent an improvement compared to MVCT. The generated images display reduced noise and improved contrast (Figure 1). Similar findings were observed for all four metrics when evaluating both sexes using the model trained on women.

Figure 1: Result of neural network