Multimodality MRI synchronous construction based deep learning framework for MRI-guided radiotherapy synthetic CT...

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Artificial Intelligence in Medicine

Background:

• Accurate electron density information is crucial for radiation dose calculation in MRI-guided radiation therapy (MRIgRT).
• Synthesizing computed tomography (CT) images from magnetic resonance imaging (MRI) data is a key step, but requires multiple MRI modalities, which is clinically expensive and time-consuming.

Purpose of the Study:

• To develop a deep learning framework for generating synthetic CT (sCT) images from a single T1-weighted (T1) MRI image for MRIgRT.
• To overcome the clinical cost and time constraints associated with acquiring multiple MRI modalities.

Main Methods:

• A generative adversarial network (GAN) based framework was proposed, featuring a multitask generator and a multibranch discriminator.
• The generator includes a shared encoder and a splitted multibranch decoder with attention modules for feature representation and fusion.
• The framework performs synchronous construction of multimodality MRI and sCT image generation from a single T1 MRI.

Main Results:

• The proposed network achieved superior performance compared to state-of-the-art methods, demonstrating the lowest Mean Absolute Error (MAE) and Normalized Root Mean Square Error (NRMSE).
• The generated sCT images showed comparable or superior quality (PSNR, SSIM) to those generated using multiple MRI modalities.
• The framework successfully generated high-quality sCT images using only a single T1 MRI input.

Conclusions:

• The proposed deep learning framework provides an effective and economical solution for sCT image generation in clinical MRIgRT.
• This single-input approach significantly reduces the cost and time associated with acquiring multiple MRI modalities.
• The method enables more accessible and efficient treatment planning for MRI-guided radiation therapy.