Corrigendum to “Proton dose calculation on cone-beam computed tomography using unsupervised 3D deep learning networks”. [Phys Imaging Radiat Oncol 2024;32:100658] (Physics and Imaging in Radiation Oncology (2024) 32, (S2405631624001283), (10.1016/j.phro.2024.100658))

The authors would like to point out that a mistake was made in the calculation of the Structural Similarity Index Measure (SSIM). In the published version, the constants c ₁ and c ₂ were calculated using a dynamic range of L = 65,535 (corresponding to 16-bit images). However, this value is incorrect, as the actual data range of the CBCT, CT, and synthetic CT images used in the SSIM calculation was limited to [−1000; 3071]. Hence, the correct dynamic range for the SSIM should have been L = 4071. A new Table 1 with the corrected SSIM values is shown below. Additionally, the corrected explanatory text to Table 1 (in Section 3.1 in the main article) is: “All three sCT networks substantially improved the PSNR, SSIM, ME, and MAE for all four structures compared to the CBCT (Table 1). The CycleCUT network had a significantly (p ≤ 0.05) better PSNR, SSIM, and MAE for the target; PSNR, SSIM, MAE, and ME for the mandible; and MAE and ME for the body, while the CycleGAN had a significantly (p ≤ 0.05) better ME for the oral cavity.” The correction to the SSIM values does not alter the conclusions of the paper; rather, it further strengthens the demonstrated quality of the synthetic CT images compared to the CBCT images, particularly for the CycleCUT network. The authors apologize for this oversight.