Deep Learning with Domain Adaptation for Accelerated Projection-Reconstruction MR (1703.01135v2)

Abstract: Purpose: The radial k-space trajectory is a well-established sampling trajectory used in conjunction with magnetic resonance imaging. However, the radial k-space trajectory requires a large number of radial lines for high-resolution reconstruction. Increasing the number of radial lines causes longer acquisition time, making it more difficult for routine clinical use. On the other hand, if we reduce the number of radial lines, streaking artifact patterns are unavoidable. To solve this problem, we propose a novel deep learning approach with domain adaptation to restore high-resolution MR images from under-sampled k-space data. Methods: The proposed deep network removes the streaking artifacts from the artifact corrupted images. To address the situation given the limited available data, we propose a domain adaptation scheme that employs a pre-trained network using a large number of x-ray computed tomography (CT) or synthesized radial MR datasets, which is then fine-tuned with only a few radial MR datasets. Results: The proposed method outperforms existing compressed sensing algorithms, such as the total variation and PR-FOCUSS methods. In addition, the calculation time is several orders of magnitude faster than the total variation and PR-FOCUSS methods.Moreover, we found that pre-training using CT or MR data from similar organ data is more important than pre-training using data from the same modality for different organ. Conclusion: We demonstrate the possibility of a domain-adaptation when only a limited amount of MR data is available. The proposed method surpasses the existing compressed sensing algorithms in terms of the image quality and computation time.

• The paper proposes a deep learning framework with domain adaptation to accelerate projection-reconstruction MR by removing streaking artifacts from under-sampled data.
• The method uses a neural network pre-trained on CT or synthesized data and fine-tuned on limited MR data, achieving better image quality and significantly faster computation than traditional algorithms.
• This approach has practical implications for clinical MRI by enabling faster, high-quality imaging with reduced data and demonstrates the potential of leveraging cross-modality data for reconstruction.

Deep Learning with Domain Adaptation for Accelerated Projection-Reconstruction MR

The paper "Deep Learning with Domain Adaptation for Accelerated Projection-Reconstruction MR" presents an innovative approach to enhancing the process of magnetic resonance imaging (MRI) reconstruction by utilizing deep learning techniques. This research focuses on addressing the challenges encountered with the radial k-space trajectory used in MRIs. The primary issue with this trajectory lies in its requirement for numerous radial lines to achieve high-resolution imaging, which consequentially results in longer acquisition times. The paper proposes a solution utilizing a novel deep learning framework combined with a domain adaptation approach to mitigate streaking artifact patterns ensuing from reduced radial line sampling.

Methodology

The proposed solution involves a deep neural network designed to remove streaking artifacts from under-sampled MRI k-space data. The key innovation in this network is its domain adaptation strategy, which leverages a pre-trained model initially trained on x-ray computed tomography (CT) or synthesized radial MR datasets. This pre-trained network is fine-tuned with a limited set of radial MR data, facilitating superior reconstruction despite the scarcity of MR-specific datasets. This approach effectively reduces computation times significantly, with performance improvements observed over traditional methods like total variation and PR-FOCUSS algorithms.

Results

The numerical results detailed in the paper underscore the efficiency of the proposed deep learning architecture. The network surpasses existing compressed sensing algorithms in performance by providing higher quality image reconstructions and achieving significantly faster computation times. Specifically, the practical advantages of this framework are evident in its capacity to restore high-resolution images effectively from minimal k-space data, thereby making it more adaptable for routine clinical MRI usage.

Implications and Future Directions

The implications of this work are substantial both theoretically and practically. By successfully applying domain adaptation in the context of MRI, this research opens avenues for utilizing cross-modality data in deep learning-based medical image reconstruction. The findings suggest that the similarities between projection-reconstruction MR and CT can be harnessed effectively to improve MR imaging processes, particularly when data availability is constrained. Future research could explore extending this technique to different MR modalities and potentially tailoring the domain adaptation methodology for diverse medical imaging systems beyond MR and CT. Additionally, investigating the applicability of this approach in dynamic MRI contexts and other imaging scenarios could yield further enhancements in medical imaging technologies.

In conclusion, this paper presents a significant contribution to the field of medical image reconstruction, highlighting the potential of deep learning models enhanced by domain adaptation. This approach not only improves image quality and reduces computation times but also demonstrates the feasibility of leveraging inter-modality data for optimized performance in medical imaging systems.