CT synthesis from MRI using multi-cycle GAN for head-and-neck radiation therapy

Area of Science:

• Medical Imaging
• Radiotherapy
• Artificial Intelligence

Background:

• Magnetic Resonance Imaging (MRI) guided Radiation Therapy planning requires synthetic Computed Tomography (sCT) generation from MRI data.
• Current image-to-image translation methods struggle to produce high-quality medical images for this application.

Purpose of the Study:

• To propose a novel framework, Multi-Cycle GAN, for high-quality synthetic CT image generation from MRI.
• To improve the accuracy of anatomical details in synthetic CT images.

Main Methods:

• Developed a novel framework named Multi-Cycle GAN.
• Incorporated a Pseudo-Cycle Consistent module for generation control.
• Utilized a domain control module for additional constraints.
• Designed a new generator, Z-Net, to enhance anatomical detail accuracy.

Main Results:

• Multi-Cycle GAN demonstrated superior performance compared to state-of-the-art methods like Cycle GAN.
• Achieved a Mean Absolute Error (MAE) of 0.0416.
• Achieved a Mean Error (ME) of 0.0340.
• Achieved a Peak Signal-to-Noise Ratio (PSNR) of 39.1053.

Conclusions:

• Multi-Cycle GAN offers a significant advancement in synthetic CT generation for MRI-guided radiotherapy.
• The proposed framework effectively improves image quality and anatomical accuracy.
• This method holds promise for enhancing radiotherapy planning accuracy and patient outcomes.