Deep learning-based convolutional neural network for intramodality brain MRI synthesis.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neuro-oncology

Background:

• Multicontrast magnetic resonance (MR) images enhance brain cancer diagnosis and treatment.
• Acquiring complete sets of MR contrasts is often not feasible in clinical practice.

Purpose of the Study:

• To develop a deep learning convolutional neural network (CNN) for image-to-image translation across standard brain MRI contrasts.
• To generate synthetic MR images from existing contrasts to augment available clinical data.

Main Methods:

• Utilized the BRATS'2018 dataset (477 patients) with T1, T2, and FLAIR contrasts.
• Developed and trained a U-Net model for nonlinear mapping between MRI contrasts.
• Evaluated synthetic images using Mean Squared Error (MSE), Mean Absolute Error (MAE), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index (SSIM).

Main Results:

• Generated synthetic MR images were nearly indistinguishable from real images across tested translations.
• Achieved average PSNR (29.44-33.25 dB), MSE (0.0005-0.0012), MAE (0.0086-0.0149), and SSIM (0.932-0.946).
• Performance was comparable to state-of-the-art deep learning models on BRATS datasets.

Conclusions:

• The U-Net model accurately performs image-to-image translation across brain MRI contrasts.
• This technique shows promise for clinical application, improving decision-making and diagnosis by providing multicontrast MRIs.
• Offers a significant step towards efficiently addressing the lack of specific MR sequences without additional patient scanning.