Prostate segmentation accuracy using synthetic MRI for high-dose-rate prostate brachytherapy treatment planning

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Computed tomography (CT) and magnetic resonance imaging (MRI) are crucial for high-dose-rate (HDR) prostate brachytherapy.
• MRI is essential for prostate segmentation, but access can be limited.
• A method to generate synthetic MRI (sMRI) from CT could overcome MRI access limitations.

Purpose of the Study:

• To develop a novel generative adversarial network (GAN), PxCGAN, for generating sMRI from CT scans.
• To evaluate the image quality and prostate segmentation accuracy of the generated sMRI.
• To compare PxCGAN with existing methods like Pix2Pix and CycleGAN.

Main Methods:

• A hybrid GAN, PxCGAN, was trained on 58 paired CT-MRI datasets from HDR prostate brachytherapy patients.
• Image quality was assessed using Mean Absolute Error (MAE), Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index (SSIM) on 20 independent datasets.
• Prostate segmentation accuracy was evaluated using Dice Similarity Coefficient (DSC), Hausdorff Distance (HD), and Mean Surface Distance (MSD) against expert delineations.

Main Results:

• PxCGAN generated sMRI with enhanced soft-tissue contrast compared to CT.
• PxCGAN demonstrated superior PSNR and SSIM compared to Pix2Pix and CycleGAN (p < 0.01).
• Prostate segmentation on sMRI achieved accuracy within the inter-observer variability range of real MRI (rMRI).

Conclusions:

• PxCGAN effectively generates sMRI from CT scans, improving soft-tissue contrast for prostate imaging.
• The accuracy of prostate segmentation using sMRI is comparable to that achieved with real MRI, even considering inter-observer variability.
• This GAN-based approach offers a viable solution for accurate prostate segmentation in HDR brachytherapy when MRI access is limited.