Real-time Cardiovascular MR with Spatio-temporal Artifact Suppression using Deep Learning - Proof of Concept in Congenital Heart Disease (1803.05192v3)

Abstract: PURPOSE: Real-time assessment of ventricular volumes requires high acceleration factors. Residual convolutional neural networks (CNN) have shown potential for removing artifacts caused by data undersampling. In this study we investigated the effect of different radial sampling patterns on the accuracy of a CNN. We also acquired actual real-time undersampled radial data in patients with congenital heart disease (CHD), and compare CNN reconstruction to Compressed Sensing (CS). METHODS: A 3D (2D plus time) CNN architecture was developed, and trained using 2276 gold-standard paired 3D data sets, with 14x radial undersampling. Four sampling schemes were tested, using 169 previously unseen 3D 'synthetic' test data sets. Actual real-time tiny Golden Angle (tGA) radial SSFP data was acquired in 10 new patients (122 3D data sets), and reconstructed using the 3D CNN as well as a CS algorithm; GRASP. RESULTS: Sampling pattern was shown to be important for image quality, and accurate visualisation of cardiac structures. For actual real-time data, overall reconstruction time with CNN (including creation of aliased images) was shown to be more than 5x faster than GRASP. Additionally, CNN image quality and accuracy of biventricular volumes was observed to be superior to GRASP for the same raw data. CONCLUSION: This paper has demonstrated the potential for the use of a 3D CNN for deep de-aliasing of real-time radial data, within the clinical setting. Clinical measures of ventricular volumes using real-time data with CNN reconstruction are not statistically significantly different from the gold-standard, cardiac gated, BH techniques.

• The paper introduces a residual U-Net deep learning method trained on synthetic data to reconstruct highly undersampled real-time cardiovascular MR images.
• CNN-based reconstruction was over five times faster than conventional methods and achieved lower RMSE and higher SSIM in synthetic data tests.
• This real-time reconstruction method is potentially useful for patients unable to breath-hold and offers a faster clinical workflow, paving the way for future AI integration in diagnostics.

Real-time Cardiovascular MR with Spatio-temporal Artifact Suppression using Deep Learning: An Analytical Examination

Overview

The research paper titled "Real-time Cardiovascular MR with Spatio-temporal Artifact Suppression using Deep Learning - Proof of Concept in Congenital Heart Disease" presents a detailed investigation into leveraging convolutional neural networks (CNN) for real-time reconstruction of highly undersampled cardiovascular magnetic resonance (MR) data. This paper specifically targets data acquired from patients with congenital heart disease (CHD), addressing challenges posed by artifacts arising from data undersampling. Residual CNNs are proposed as an alternative to conventional compressed sensing (CS) techniques, promising improvements in reconstruction speed and image quality.

Methodological Insights

The authors developed a novel deep artifact suppression algorithm grounded in a modified residual U-Net architecture. Training was performed using synthetic data derived from previously acquired breath-hold cine images from a significant cohort of 250 CHD patients. An undersampling strategy utilizing tiny Golden Angle radial spokes was employed, aiming to reduce aliasing effects in real-time image acquisition.

Key methodological aspects include:

• Training Data Preparation: Ground truth magnitude images were synthetically generated by converting breath-hold cine images to mimic real-time acquisition parameters, offering consistent spatial and temporal resolution.
• Sampling Strategy Assessment: Four radial sampling patterns were compared to assess the impact on artifact suppression, highlighting the superiority of a continuously rotating tiny golden angle (tGArot) strategy in producing noise-like aliasing conducive to effective reconstruction.
• Network Architecture and Training: A residual U-Net architecture was leveraged, employing a multi-scale decomposition of input images, trained using MSE loss to minimize reconstruction errors.

Strong Numerical Results

The paper demonstrated that CNN-based reconstruction reduced overall processing time significantly, achieving speeds over five times faster than CS. The residual U-Net trained using synthetic data achieved lower RMSE and higher SSIM compared to traditional methods in synthetic test data evaluation, reaffirming the technique's efficacy in artifact removal.

Implications and Future Developments

Practically, this research offers a viable pathway for improving clinical MR imaging, particularly for patients unable to perform breath-holds. The significant acceleration of reconstruction times introduces potential for real-time clinical application, reducing workflow bottlenecks. Theoretically, these findings underscore the power of leveraging deep learning frameworks in automating complex image reconstruction tasks, suggesting further exploration of CNN models tailored for diverse MR imaging challenges.

Future developments could focus on expanding generalization capabilities by incorporating varied respiratory patterns in training datasets, enhancing robustness across different clinical scenarios. Additionally, integrating phase information or exploring alternative loss functions may refine image fidelity. The potential role of artificial intelligence in streamlining clinical imaging and contributing to personalized diagnostics appears promising.

Conclusion

This research substantiates the role of CNNs in enhancing real-time MR image reconstruction, offering tangible benefits in speed and image quality over conventional methods. As deep learning continues to redefine imaging standards, this paper serves as a foundational step toward transitioning from traditional reconstruction paradigms to AI-enhanced imaging in the clinical field. The endeavor to refine these techniques could significantly impact the future of diagnostic medicine, particularly in challenging patient groups such as those with congenital heart disease.