Computed tomography synthesis from magnetic resonance imaging using cycle Generative Adversarial Networks with multicenter learning

Abstract

Background and Purpose: Addressing the need for accurate dose calculation in MRI-only radiotherapy, the generation of synthetic Computed Tomography (sCT) from MRI has emerged. Deep learning (DL) techniques, have shown promising results in achieving high sCT accuracies. However, existing sCT synthesis methods are often center-specific, posing a challenge to their generalizability. To overcome this limitation, recent studies have proposed approaches, such as multicenter training . Material and methods: The purpose of this work was to propose a multicenter sCT synthesis by DL, using a 2D cycle-GAN on 128 prostate cancer patients, from four different centers. Four cases were compared: monocenter cases, monocenter training and test on another center, multicenter trainings and a test on a center not included in the training and multicenter trainings with an included center in the test. Trainings were performed using 20 patients. sCT accuracy evaluation was performed using Mean Absolute Error, Mean Error and Peak-Signal-to-Noise-Ratio. Dose accuracy was assessed with gamma index and Dose Volume Histogram comparison. Results: Qualitative, quantitative and dose results show that the accuracy of sCTs for monocenter trainings and multicenter trainings using a seen center in the test did not differ significantly. However, when the test involved an unseen center, the sCT quality was inferior. Conclusions: The aim of this work was to propose generalizable multicenter training for MR-to-CT synthesis. It was shown that only a few data from one center included in the training cohort allows sCT accuracy equivalent to a monocenter study.

Keywords: Magnetic resonance imaging; Multicenter; Radiotherapy treatment planning; Synthetic-CT; cycle-GAN.

Fig. 1

Workflow of the sCT generation from multicenter cohort.

Fig. 2

Cycle-GAN model adapted from . Generator $G_A$ converts MR images into sCT or sMR into sCT, $G_B$ does the opposite. The perceptual loss as $L_{\mathit{GA}}$ and $L_{\mathit{GB}}$ is computed, it compares the sCT (resp sMR) with the reference CT (resp MR). At each iteration, the cycle loss ($L_C$) compares the real CT (respectively MR) with the sCT (resp. sMR) obtained with the sMR (resp. the sCT). Finally, the BCE loss function $L_{\mathit{DA}}$ (resp. the $L_{\mathit{DB}}$) is computed and determines the probability of having a real CT (resp. real MR).

Fig. 3

Preprocessed CT and MRI and image and sCT results according to the test dataset and the case: case A) monocenter study, case B) monocenter training using unseen dataset in the test, case C) multicenter training using unseen data in the test, and case D) multicenter training using seen data in the test.

Fig. 4

Results of sCT dose evaluation. Section 1 shows the gamma pass rate results according to the test dataset (D1-D4) and the case (A,B,C or D). Sections 2, 3, 4 presents absolute dose differences. Section 2, presents the results of the absolute differences between the $D_{95}\%$ of the dose on reference CT and the $D_{95}\%$ of calculated dose on sCT in the prostate according to the test dataset and the case. Section 3 presents the results of the absolute differences between the $D_{\mathit{mean}}\%$ calculated on the reference CT and the $D_{\mathit{mean}}\%$calculated on sCT in the rectum according to the test dataset and the case. Section 4 presents the same results as Section 3 but in the bones.