Compressed Sensing MRI Reconstruction using a Generative Adversarial Network with a Cyclic Loss (1709.00753v2)

Abstract: Compressed Sensing MRI (CS-MRI) has provided theoretical foundations upon which the time-consuming MRI acquisition process can be accelerated. However, it primarily relies on iterative numerical solvers which still hinders their adaptation in time-critical applications. In addition, recent advances in deep neural networks have shown their potential in computer vision and image processing, but their adaptation to MRI reconstruction is still in an early stage. In this paper, we propose a novel deep learning-based generative adversarial model, RefineGAN, for fast and accurate CS-MRI reconstruction. The proposed model is a variant of fully-residual convolutional autoencoder and generative adversarial networks (GANs), specifically designed for CS-MRI formulation; it employs deeper generator and discriminator networks with cyclic data consistency loss for faithful interpolation in the given under-sampled k-space data. In addition, our solution leverages a chained network to further enhance the reconstruction quality. RefineGAN is fast and accurate -- the reconstruction process is extremely rapid, as low as tens of milliseconds for reconstruction of a 256x256 image, because it is one-way deployment on a feed-forward network, and the image quality is superior even for extremely low sampling rate (as low as 10%) due to the data-driven nature of the method. We demonstrate that RefineGAN outperforms the state-of-the-art CS-MRI methods by a large margin in terms of both running time and image quality via evaluation using several open-source MRI databases.

Compressed Sensing MRI Reconstruction Using a Generative Adversarial Network with a Cyclic Loss

This paper presents a novel approach to Compressed Sensing MRI reconstruction leveraging a Generative Adversarial Network (GAN) framework named RefineGAN. The primary motivation behind this development arises from the need to enhance MRI imaging speed without sacrificing image quality, addressing a critical bottleneck in time-sensitive medical diagnostics.

Methodological Innovation

The proposed RefineGAN architecture represents a significant advancement in deep learning-based MRI reconstructions. It combines elements of convolutional autoencoders and residual networks, integrating GANs with a cyclic data consistency loss. This cyclic loss plays a crucial role by maintaining consistency between under-sampled input and fully reconstructed images, improving interpolation of undersampled $k$-space data. The architecture emphasizes the utilization of deeper generative and discriminative networks, thus enhancing both the speed and accuracy of MRI reconstruction.

Performance Evaluation

Quantitative evaluations show that RefineGAN significantly outperforms traditional CS-MRI methods. The reported reconstruction time is as low as tens of milliseconds for a 256x256 image, a substantial improvement that aligns with the operational needs of clinical settings. This efficiency stems from the deployment of a single-pass feed-forward network, eliminating the iterative processes typical in conventional CS-MRI techniques. The method captures superior image quality, even at low sampling rates (down to 10%), surpassing other state-of-the-art approaches, as indicated by notable increases in PSNR, SSIM, and NRMSE metrics.

Implications and Future Directions

The results suggest that RefineGAN has significant practical implementations in clinical radiology, especially where rapid and accurate diagnostic imaging is paramount. Beyond practical implications, it also contributes theoretically to the application of GANs in medical imaging, demonstrating how cyclic consistency loss can be effectively used to maintain data fidelity in severely undersampled conditions.

Future research directions could explore extensions to dynamic MRI, demanding the adaptation of the current framework to handle temporal variations in image sequences. Additionally, scaling RefineGAN for complex-valued MRI datasets offers a promising research avenue, potentially enhancing its applicability in diverse clinical scenarios.

In conclusion, RefineGAN represents a robust advancement in the application of GANs to CS-MRI, offering a rapid, high-quality alternative to existing methodologies. Its development highlights the potential of integrating advanced deep network architectures to address longstanding challenges in medical imaging.