Abstract:Generative adversarial networks (GANs) have shown remarkable success in generating realistic images and are increasingly used in medical imaging for image-to-image translation tasks. However, GANs tend to suffer from a frequency bias towards low frequencies, which can lead to the removal of important structures in the generated images. To address this issue, we propose a novel frequency-aware image-to-image translation framework based on the supervised RegGAN approach, which we call fRegGAN. The framework employs a K-space loss to regularize the frequency content of the generated images and incorporates well-known properties of MRI K-space geometry to guide the network training process. By combine our method with the RegGAN approach, we can mitigate the effect of training with misaligned data and frequency bias at the same time. We evaluate our method on the public BraTS dataset and outperform the baseline methods in terms of both quantitative and qualitative metrics when synthesizing T2-weighted from T1-weighted MR images. Detailed ablation studies are provided to understand the effect of each modification on the final performance. The proposed method is a step towards improving the performance of image-to-image translation and synthesis in the medical domain and shows promise for other applications in the field of image processing and generation.

What problem does this paper attempt to address?

The paper primarily addresses the issue of frequency bias in Generative Adversarial Networks (GANs) for medical image translation tasks and proposes a novel solution. Specifically, since GANs tend to favor low-frequency information, this can lead to the loss of important structures in the generated medical images, especially in applications where high-frequency details (such as edges) need to be preserved. To solve this problem, the authors propose a new frequency-aware image-to-image translation framework based on a supervised RegGAN method, called fRegGAN. The main contributions of fRegGAN include: 1. **Improved RegGAN architecture**: By making universally applicable modifications to the architecture and training process to further explore the RegGAN method, and extending its CycleGAN method by adding a second registration network to enhance the performance of both generators. 2. **Introduction of K-space loss**: Utilizing constraints in the K-space (the frequency domain representation of MRI data) to regularize and guide the network training process. This regularization technique is inspired by MRI acquisition and reconstruction methods that use the distribution of image feature information in K-space to reduce noise or accelerate image acquisition. 3. **Evaluation and results**: The proposed fRegGAN method was evaluated on the publicly available BraTS dataset, and the results showed that the method outperformed baseline methods (such as RegGAN and CycleGAN) in both quantitative and qualitative metrics. Detailed ablation studies were conducted to understand the impact of each modification on the final performance. In summary, the paper aims to improve the performance of GANs in medical image translation tasks by combining frequency domain information with the existing RegGAN framework, particularly by increasing the retention of high-frequency information, thereby enhancing the quality of generated images, which is of significant importance for clinical applications.

fRegGAN with K-space Loss Regularization for Medical Image Translation

Frequency Domain Decomposition Translation for Enhanced Medical Image Translation Using GANs

MedGAN: Medical Image Translation using GANs

TarGAN: Target-Aware Generative Adversarial Networks for Multi-modality Medical Image Translation

Uncertainty-Guided Progressive GANs for Medical Image Translation

AV-GAN: Attention-Based Varifocal Generative Adversarial Network for Uneven Medical Image Translation

ReeGAN: MRI image edge-preserving synthesis based on GANs trained with misaligned data

Breaking the Dilemma of Medical Image-to-image Translation

Memory-efficient GAN-based domain translation of high resolution 3D medical images

GAN-GA: A Generative Model based on Genetic Algorithm for Medical Image Generation

Segmentation-Renormalized Deep Feature Modulation for Unpaired Image Harmonization

BioGAN: An unpaired GAN-based image to image translation model for microbiological images

A Domain Translation Framework with an Adversarial Denoising Diffusion Model to Generate Synthetic Datasets of Echocardiography Images

MRI Cross-Modality NeuroImage-to-NeuroImage Translation

Unsupervised Medical Image Translation Using Cycle-MedGAN

Cross-modality image translation of 3 Tesla Magnetic Resonance Imaging to 7 Tesla using Generative Adversarial Networks

CycleGAN Models for MRI Image Translation

Multi-modality MRI Arbitrary Transformation Using Unified Generative Adversarial Networks

Generating 3D Brain Tumor Regions in MRI using Vector-Quantization Generative Adversarial Networks

FedMed-GAN: Federated Domain Translation on Unsupervised Cross-Modality Brain Image Synthesis

Multi-Texture GAN: Exploring the Multi-Scale Texture Translation for Brain MR Images