FDG-PET to T1 Weighted MRI Translation with 3D Elicit Generative Adversarial Network (E-GAN)

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Deep learning models require substantial data for training, which is challenging to obtain for medical imaging due to privacy and collection difficulties.
• Synthesizing 3D medical images, like MRI, is complex but crucial for overcoming data limitations in computer-aided diagnosis (CAD).

Purpose of the Study:

• To generate 3D T1-weighted MRI from FDG-PET data using a novel generative adversarial network.
• To address the challenge of limited and imbalanced datasets in medical image analysis.

Main Methods:

• Proposes a separable convolution-based Elicit generative adversarial network (E-GAN).
• Reconstructs 3D T1-weighted MRI from 2D high-level features and geometrical information from a Sobel filter.
• Utilizes the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset for experiments.

Main Results:

• The E-GAN model demonstrates improved image quality compared to state-of-the-art methods on ADNI datasets.
• Achieved significant improvements in structural information: 13.73% for Peak Signal-to-Noise Ratio (PSNR) and 22.95% for Structural Similarity Index Measure (SSIM) compared to Pix2Pix GAN.
• Showcased enhanced textural information with a 6.9% improvement in homogeneity error using Haralick features compared to Pix2Pix GAN.

Conclusions:

• The proposed E-GAN effectively synthesizes 3D T1-weighted MRI from FDG-PET, offering a viable solution for data augmentation in medical imaging.
• The method significantly enhances both structural and textural fidelity of synthesized images, outperforming existing GAN-based approaches.