Generalising Deep Learning MRI Reconstruction across Different Domains

Ouyang, Cheng; Schlemper, Jo; Biffi, Carlo; Seegoolam, Gavin; Caballero, José; Price, Anthony N.; Hajnal, Joseph V.; Rueckert, Daniel

doi:10.48550/arxiv.1902.10815

Cited by 2 publications

(3 citation statements)

References 3 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This indicates that the learned base of the universal model captures the anatomy-common knowledge and using ASPIN further allows adaptation to a new anatomy even with small-scale data. The effectiveness of the large-scale data is also seen in [3,5,16]. We visualized the reconstructed images of different anatomies under 6× acceleration in Fig.…”

Section: Algorithm Comparisonmentioning

confidence: 87%

“…First, numerous models lead to a large number of parameters, which makes it hard to deploy on commercial MRI machines.Second, datasets of various anatomies contain shared prior knowledge [14]; separately trained networks do not exploit such common knowledge, which may limit the effectiveness of the trained models. While authors in [16] attempted to generalize DL-based reconstruction, the network is trained purely on natural images with limited anatomical knowledge. Therefore, it is highly desirable to train a universal model that exploits cross-anatomy common knowledge and conducts reconstruction for various datasets, even when some datasets are very small.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Universal Undersampled MRI Reconstruction

Liu¹,

Wang²,

Liu³

et al. 2021

Preprint

View full text Add to dashboard Cite

Deep neural networks have been extensively studied for undersampled MRI reconstruction. While achieving state-of-the-art performance, they are trained and deployed specifically for one anatomy with limited generalization ability to another anatomy. Rather than building multiple models, a universal model that reconstructs images across different anatomies is highly desirable for efficient deployment and better generalization. Simply mixing images from multiple anatomies for training a single network does not lead to an ideal universal model due to the statistical shift among datasets of various anatomies, the need to retrain from scratch on all datasets with the addition of a new dataset, and the difficulty in dealing with imbalanced sampling when the new dataset is further of a smaller size. In this paper, for the first time, we propose a framework to learn a universal deep neural network for undersampled MRI reconstruction. Specifically, anatomy-specific instance normalization is proposed to compensate for statistical shift and allow easy generalization to new datasets. Moreover, the universal model is trained by distilling knowledge from available independent models to further exploit representations across anatomies. Experimental results show the proposed universal model can reconstruct both brain and knee images with high image quality. Also, it is easy to adapt the trained model to new datasets of smaller size, i.e., abdomen, cardiac and prostate, with little effort and superior performance.

show abstract

Section: Algorithm Comparisonmentioning

confidence: 87%

Section: Introductionmentioning

confidence: 99%

Universal Undersampled MRI Reconstruction

Liu¹,

Wang²,

Liu³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Ledig conducted a study using a generative adversarial network (GAN) to generate virtual data for SR using arbitrary random numbers between generators and discriminators [13][14][15][16][17][18][19][20]. This research also used a super-resolution GAN (SRGAN), whereas the conventional method uses mean squared reconstruction error to obtain the peak signal-to-noise ratio (PSNR).…”

Section: Introductionmentioning

confidence: 99%

Robust Pose Estimation of Pedestrians with a Deep Neural Networks

Yi,

Cho

2023

Int. J. Adv. Sci. Eng. Inf. Technol.

View full text Add to dashboard Cite

In this paper, we provide a method for robust estimation of pedestrian pose that is especially useful for autonomous vehicles traveling toward pedestrians far away. Pedestrians in the far distance appear relatively small when seen by a camera, making it difficult to estimate the pedestrian's pose. We use fused deep neural networks (DNNs) to resolve the problems presented by pedestrians in the far distance. First, DNNs are used to detect pedestrians and enlarge the observed image. Next, the DNN method of pose estimation is applied. The proposed method uses a single camera to estimate the posture of a pedestrian in the far distance. Far-off pedestrians observed by cameras in moving cars appear as low-resolution images of non-rigid bodies. Detection and orientation estimation are difficult with conventional image processing methods. We used a series of DNNs to detect pedestrians, improve data availability, and estimate challenging postures to address these limitations. In this paper, we propose a method based on the multi-stage fusion of DNNs to solve a difficult problem for a single DNN. The experimental results established the superiority of the proposed method when applied to data challenging for conventional pose estimation methods. Applications of the proposed method include observing small objects and objects in the far distance. The method may be especially useful in surveillance systems, sports broadcasting, and other applications requiring human posture estimation.

show abstract

Generalising Deep Learning MRI Reconstruction across Different Domains

Cited by 2 publications

References 3 publications

Universal Undersampled MRI Reconstruction

Universal Undersampled MRI Reconstruction

Robust Pose Estimation of Pedestrians with a Deep Neural Networks

Contact Info

Product

Resources

About