Undersampling artifact reduction for free-breathing 3D stack-of-radial MRI based on a deep adversarial learning network.

PURPOSE: Stack-of-radial MRI allows free-breathing abdominal scans, however, it requires relatively long acquisition time. Undersampling reduces scan time but can cause streaking artifacts and degrade image quality. This study developed deep learning networks with adversarial loss and evaluated the performance of reducing streaking artifacts and preserving perceptual image sharpness. METHODS: A 3D generative adversarial network (GAN) was developed for reducing streaking artifacts in stack-of-radial abdominal scans. Training and validation datasets were self-gated to 5 respiratory states to reduce motion artifacts and to effectively augment the data. The network used a combination of three loss functions to constrain the anatomy and preserve image quality: adversarial loss, mean-squared-error loss and structural similarity index loss. The performance of the network was investigated for 3-5 times undersampled data from 2 institutions. The performance of the GAN for 5 times accelerated images was compared with a 3D U-Net and evaluated using quantitative NMSE, SSIM and region of interest (ROI) measurements as well as qualitative scores of radiologists. RESULTS: The 3D GAN showed similar NMSE (0.0657 vs. 0.0559, p = 0.5217) and significantly higher SSIM (0.841 vs. 0.798, p < 0.0001) compared to U-Net. ROI analysis showed GAN removed streaks in both the background air and the tissue and was not significantly different from the reference mean and variations. Radiologists' scores showed GAN had a significant improvement of 1.6 point (p = 0.004) on a 4-point scale in streaking score while no significant difference in sharpness score compared to the input. CONCLUSION: 3D GAN removes streaking artifacts and preserves perceptual image details.

Low rank off-resonance correction for double half-echo k-space acquisitions.

The present study describes a model-based approach for correcting off-resonance in the context of double half-echo k-space acquisitions. This technique employs center-out readouts in forward and reverse directions to reduce echo-times but is sensitive to off-resonance, which manifests as pixel shifts in both directions. Demodulating the k-space signal with a constant off-resonance term per slice removes pixel shifts and results in a marked reduction in blurring. Phantom and in vivo datasets are demonstrated from low bandwidth sodium imaging.

Automatic segmentation of peripheral arteries and veins in ferumoxytol-enhanced MR angiography.

PURPOSE: To automate the segmentation of the peripheral arteries and veins in the lower extremities based on ferumoxytol-enhanced MR angiography (FE-MRA). METHODS: Our automated pipeline has 2 sequential stages. In the first stage, we used a 3D U-Net with local attention gates, which was trained based on a combination of the Focal Tversky loss with region mutual loss under a deep supervision mechanism to segment the vasculature from the high-resolution FE-MRA datasets. In the second stage, we used time-resolved images to separate the arteries from the veins. Because the ultimate segmentation quality of the arteries and veins relies on the performance of the first stage, we thoroughly evaluated the different aspects of the segmentation network and compared its performance in blood vessel segmentation with currently accepted state-of-the-art networks, including Volumetric-Net, DeepVesselNet-FCN, and Uception. RESULTS: We achieved a competitive F1 = 0.8087 and recall = 0.8410 for blood vessel segmentation compared with F1 = (0.7604, 0.7573, 0.7651) and recall = (0.7791, 0.7570, 0.7774) obtained with Volumetric-Net, DeepVesselNet-FCN, and Uception. For the artery and vein separation stage, we achieved F1 = (0.8274/0.7863) in the calf region, which is the most challenging region in peripheral arteries and veins segmentation. CONCLUSION: Our pipeline is capable of fully automatic vessel segmentation based on FE-MRA without need for human interaction in <4 min. This method improves upon manual segmentation by radiologists, which routinely takes several hours.

Temporally aware volumetric generative adversarial network-based MR image reconstruction with simultaneous respiratory motion compensation: Initial feasibility in 3D dynamic cine cardiac MRI.

PURPOSE: Develop a novel three-dimensional (3D) generative adversarial network (GAN)-based technique for simultaneous image reconstruction and respiratory motion compensation of 4D MRI. Our goal was to enable high-acceleration factors 10.7X-15.8X, while maintaining robust and diagnostic image quality superior to state-of-the-art self-gating (SG) compressed sensing wavelet (CS-WV) reconstruction at lower acceleration factors 3.5X-7.9X. METHODS: Our GAN was trained based on pixel-wise content loss functions, adversarial loss function, and a novel data-driven temporal aware loss function to maintain anatomical accuracy and temporal coherence. Besides image reconstruction, our network also performs respiratory motion compensation for free-breathing scans. A novel progressive growing-based strategy was adapted to make the training process possible for the proposed GAN-based structure. The proposed method was developed and thoroughly evaluated qualitatively and quantitatively based on 3D cardiac cine data from 42 patients. RESULTS: Our proposed method achieved significantly better scores in general image quality and image artifacts at 10.7X-15.8X acceleration than the SG CS-WV approach at 3.5X-7.9X acceleration (4.53 ± 0.540 vs. 3.13 ± 0.681 for general image quality, 4.12 ± 0.429 vs. 2.97 ± 0.434 for image artifacts, P < .05 for both). No spurious anatomical structures were observed in our images. The proposed method enabled similar cardiac-function quantification as conventional SG CS-WV. The proposed method achieved faster central processing unit-based image reconstruction (6 s/cardiac phase) than the SG CS-WV (312 s/cardiac phase). CONCLUSION: The proposed method showed promising potential for high-resolution (1 mm3 ) free-breathing 4D MR data acquisition with simultaneous respiratory motion compensation and fast reconstruction time.

Slice encoding for the reduction of outflow signal artifacts in cine balanced SSFP imaging.

PURPOSE: Standard balanced SSFP (bSSFP) cine MRI often suffers from blood outflow artifacts. We propose a method that spatially encodes these outflowing spins to reduce their effects in the intended slice. METHODS: Bloch simulations were performed to characterize through-plane flow and to investigate how the use of phase encoding along the slice select's direction ("slice encoding") could alleviate its issues. Phantom scans and in vivo cines were acquired on a 3T system, comparing the standard 2D acquisition to the proposed slice-encoding method. Nineteen healthy volunteers were recruited for short-axis and horizontal long-axis oriented scans. An expert radiologist evaluated each slice-encoded/standard cine pairs in a rank comparison test and graded their quality on a 1-5 scale. The grades were used for a nonparametric paired evaluation for independent samples with a null hypothesis that there was no statistical difference between the two quality-grade distributions for α = 0.05 significance. RESULTS: Bloch simulation results demonstrated this technique's feasibility, showing a fully resolved slice profile given a sufficient number of slice encodes. These results were confirmed with the phantom experiments. Each in vivo slice-encoded cine had a higher quality than its corresponding standard acquisition. The nonparametric paired evaluation came to 0.01 significance, encouraging us to reject the null hypothesis and conclude that slice-encoding effectively works in reducing outflow effects. CONCLUSION: The slice-encoding balanced SSFP technique is helpful in mitigating outflow effects and is achievable within a single breath hold, being a useful alternative for cases in which the flow artifacts are significant.

Prediction of soft tissue sarcoma response to radiotherapy using longitudinal diffusion MRI and a deep neural network with generative adversarial network-based data augmentation.

PURPOSE: The goal of this study was to predict soft tissue sarcoma response to radiotherapy (RT) using longitudinal diffusion-weighted MRI (DWI). A novel deep-learning prediction framework along with generative adversarial network (GAN)-based data augmentation was investigated for the response prediction. METHODS: Thirty soft tissue sarcoma patients who were treated with five-fraction hypofractionated radiation therapy (RT, 6Gy×5) underwent diffusion-weighted MRI three times throughout the RT course using an MR-guided radiotherapy system. Pathologic treatment effect (TE) scores, ranging from 0-100%, were obtained from the post-RT surgical specimen as a surrogate of patient treatment response. Patients were divided into three classes based on the TE score (TE ≤ 20%, 20% < TE < 90%, TE ≥ 90%). Apparent diffusion coefficient (ADC) maps of the tumor from the three time points were combined as 3-channel images. An auxiliary classifier generative adversarial network (ACGAN) was trained on 20 patients to augment the data size. A total of 15,000 synthetic images were generated for each class. A prediction model based on a previously described VGG-19 network was trained using the synthesized data, validated on five unseen validation patients, and tested on the remaining five test patients. The entire process was repeated seven times, each time shuffling the training, validation, and testing datasets such that each patient was tested at least once during the independent test stage. Prediction performance for slice-based prediction and patient-based prediction was evaluated. RESULTS: The average training and validation accuracies were 86.5% ± 1.6% and 84.8% ± 1.8%, respectively, indicating that the generated samples were good representations of the original patient data. Among the seven rounds of testing, slice by slice prediction accuracy ranged from 81.6% to 86.8%. The overall accuracy of the independent test sets was 83.3%. For patient-based prediction, 80% was achieved in one round and 100% was achieved in the remaining six rounds. The mean accuracy was 97.1%. CONCLUSION: This study demonstrated the potential to use deep learning to predict the pathologic treatment effect from longitudinal DWI. Accuracies of 83.3% and 97.1% were achieved on independent test sets for slice-based and patient-based prediction respectively.

Minimizing echo and repetition times in magnetic resonance imaging using a double half-echo k-space acquisition and low-rank reconstruction.

Sampling k-space asymmetrically (ie, partial Fourier sampling) in the readout direction is a common way to reduce the echo time (TE) during magnetic resonance image acquisitions. This technique requires overlap around the center of k-space to provide a calibration region for reconstruction, which limits the minimum fractional echo to ~60% before artifacts are observed. The present study describes a method for reconstructing images from exact half echoes using two separate acquisitions with reversed readout polarity, effectively providing a full line of k-space without additional data around central k-space. This approach can benefit sequences or applications that prioritize short TE, short inter-echo spacing or short repetition time. An example of the latter is demonstrated to reduce banding artifacts in balanced steady-state free precession.

Retrospective respiratory motion correction in cardiac cine MRI reconstruction using adversarial autoencoder and unsupervised learning.

The aim of this study was to develop a deep neural network for respiratory motion compensation in free-breathing cine MRI and evaluate its performance. An adversarial autoencoder network was trained using unpaired training data from healthy volunteers and patients who underwent clinically indicated cardiac MRI examinations. A U-net structure was used for the encoder and decoder parts of the network and the code space was regularized by an adversarial objective. The autoencoder learns the identity map for the free-breathing motion-corrupted images and preserves the structural content of the images, while the discriminator, which interacts with the output of the encoder, forces the encoder to remove motion artifacts. The network was first evaluated based on data that were artificially corrupted with simulated rigid motion with regard to motion-correction accuracy and the presence of any artificially created structures. Subsequently, to demonstrate the feasibility of the proposed approach in vivo, our network was trained on respiratory motion-corrupted images in an unpaired manner and was tested on volunteer and patient data. In the simulation study, mean structural similarity index scores for the synthesized motion-corrupted images and motion-corrected images were 0.76 and 0.93 (out of 1), respectively. The proposed method increased the Tenengrad focus measure of the motion-corrupted images by 12% in the simulation study and by 7% in the in vivo study. The average overall subjective image quality scores for the motion-corrupted images, motion-corrected images and breath-held images were 2.5, 3.5 and 4.1 (out of 5.0), respectively. Nonparametric-paired comparisons showed that there was significant difference between the image quality scores of the motion-corrupted and breath-held images (P < .05); however, after correction there was no significant difference between the image quality scores of the motion-corrected and breath-held images. This feasibility study demonstrates the potential of an adversarial autoencoder network for correcting respiratory motion-related image artifacts without requiring paired data.

Fast and accurate calculation of myocardial T1 and T2 values using deep learning Bloch equation simulations (DeepBLESS).

PURPOSE: To propose and evaluate a deep learning model for rapid and accurate calculation of myocardial T1 /T2 values based on a previously proposed Bloch equation simulation with slice profile correction (BLESSPC) method. METHODS: Deep learning Bloch equation simulations (DeepBLESS) models are proposed for rapid and accurate T1 estimation for the MOLLI T1 mapping sequence with balanced SSFP readouts and T1 /T2 estimation for a radial simultaneous T1 and T2 mapping (radial T1 -T2 ) sequence. The DeepBLESS models were trained separately based on simulated radial T1 -T2 and MOLLI data, respectively. The DeepBLESS T1 -T2 estimation accuracy was evaluated based on simulated data with different noise levels. The DeepBLESS model was compared with BLESSPC in simulation, phantom, and in vivo studies for the MOLLI sequence at 1.5 T and radial T1 -T2 sequence at 3 T. RESULTS: After DeepBLESS was trained, in phantom studies, DeepBLESS and BLESSPC achieved similar accuracy and precision in T1 -T2 estimations for both MOLLI and radial T1 -T2 (P > .05). For in vivo, DeepBLESS and BLESSPC generated similar myocardial T1 /T2 values for radial T1 -T2 at 3 T (T1 : 1366 ± 31 ms for both methods, P > .05; T2 : 37.4 ms ± 0.9 ms for both methods, P > .05), and similar myocardial T1 values for the MOLLI sequence at 1.5 T (1044 ± 20 ms for both methods, P > .05). DeepBLESS generated a T1 /T2 map in less than 1 second. CONCLUSION: The DeepBLESS model offers an almost instantaneous approach for estimating accurate T1 /T2 values, replacing BLESSPC for both MOLLI and radial T1 -T2 sequences, and is promising for multiparametric mapping in cardiac MRI.

Constraints in estimating the proton density fat fraction.

The study evaluates four physically motivated constraints in the estimation of the proton density fat fraction (PDFF). Least squares approaches were developed for constraining the parameters in PDFF quantification based on the physics of magnetic resonance imaging. These were smooth fieldmap, smooth initial phase, nonnegative proton density and moderate R2∗ values. The constraints were evaluated in terms of their influence on the bias and standard deviation of the estimated parameters using numerical simulations and in vivo data acquired at 0.35 T. Results show that unconstrained least squares estimation is noisy and biased and that constraints can be effective at reducing both the standard deviation and bias.

MR image reconstruction using deep learning: evaluation of network structure and loss functions.

BACKGROUND: To review and evaluate approaches to convolutional neural network (CNN) reconstruction for accelerated cardiac MR imaging in the real clinical context. METHODS: Two CNN architectures, Unet and residual network (Resnet) were evaluated using quantitative and qualitative assessment by radiologist. Four different loss functions were also considered: pixel-wise (L1 and L2), patch-wise structural dissimilarity (Dssim) and feature-wise (perceptual loss). The networks were evaluated using retrospectively and prospectively under-sampled cardiac MR data. RESULTS: Based on our assessments, we find that Resnet and Unet achieve similar image quality but that former requires only 100,000 parameters compared to 1.3 million parameters for the latter. The perceptual loss function performed significantly better than L1, L2 or Dssim loss functions as determined by the radiologist scores. CONCLUSIONS: CNN image reconstruction using Resnet yields comparable image quality to Unet with 10X the number of parameters. This has implications for training with significantly lower data requirements. Network training using the perceptual loss function was found to better agree with radiologist scoring compared to L1, L2 or Dssim loss functions.

Parallel imaging and convolutional neural network combined fast MR image reconstruction: Applications in low-latency accelerated real-time imaging.

PURPOSE: To develop and evaluate a parallel imaging and convolutional neural network combined image reconstruction framework for low-latency and high-quality accelerated real-time MR imaging. METHODS: Conventional Parallel Imaging reconstruction resolved as gradient descent steps was compacted as network layers and interleaved with convolutional layers in a general convolutional neural network. All parameters of the network were determined during the offline training process, and applied to unseen data once learned. The proposed network was first evaluated for real-time cardiac imaging at 1.5 T and real-time abdominal imaging at 0.35 T, using threefold to fivefold retrospective undersampling for cardiac imaging and threefold retrospective undersampling for abdominal imaging. Then, prospective undersampling with fourfold acceleration was performed on cardiac imaging to compare the proposed method with standard clinically available GRAPPA method and the state-of-the-art L1-ESPIRiT method. RESULTS: Both retrospective and prospective evaluations confirmed that the proposed network was able to images with a lower noise level and reduced aliasing artifacts in comparison with the single-coil based and L1-ESPIRiT reconstructions for cardiac imaging at 1.5 T, and the GRAPPA and L1-ESPIRiT reconstructions for abdominal imaging at 0.35 T. Using the proposed method, each frame can be reconstructed in less than 100 ms, suggesting its clinical compatibility. CONCLUSION: The proposed Parallel Imaging and convolutional neural network combined reconstruction framework is a promising technique that allows low-latency and high-quality real-time MR imaging.