• Corpus ID: 246430767

Validation and Generalizability of Self-Supervised Image Reconstruction Methods for Undersampled MRI

  title={Validation and Generalizability of Self-Supervised Image Reconstruction Methods for Undersampled MRI},
  author={Thomas Yu and Tom Hilbert and Gian Franco Piredda and Arun A. Joseph and Gabriele Bonanno and Salim Zenkhri and Patrick Omoumi and Meritxell Bach Cuadra and Erick Jorge Canales-Rodr{\'i}guez and Thomas Kober and Jean-Philippe Thiran},
Deep learning methods have become the state of the art for undersampled MR reconstruction. Particularly for cases where it is infeasible or impossible for ground truth, fully sampled data to be acquired, self-supervised machine learning methods for reconstruction are becoming increasingly used. However potential issues in the validation of such methods, as well as their generalizability, remain underexplored. In this paper, we investigate important aspects of the validation of self-supervised… 
1 Citations

Figures and Tables from this paper

NESTANets: Stable, accurate and efficient neural networks for analysis-sparse inverse problems

This work presents a novel construction of accurate, stable and efficient neural networks for inverse problems with general analysis-sparse models, termed NESTANets, and demonstrates the construction of e-scientific neural networks based on unrolling with guaranteed stability and accuracy.



Measuring Robustness in Deep Learning Based Compressive Sensing

The robustness of different approaches for image reconstruction including trained and un-trained neural networks as well as traditional sparsity-based methods is measured and it is found that both trained andun-trained methods are vulnerable to adversarial perturbations.

Assessment of the generalization of learned image reconstruction and the potential for transfer learning

The goal of this study is to assess the influence of image contrast, SNR, and image content on the generalization of learned image reconstruction, and to demonstrate the potential for transfer learning.

Deep Image Prior

It is shown that a randomly-initialized neural network can be used as a handcrafted prior with excellent results in standard inverse problems such as denoising, super-resolution, and inpainting.

Blind image quality evaluation using perception based features

A no-reference Perception-based Image Quality Evaluator (PIQUE) for real-world imagery that attempts to quantify distortion without the need for any training data and has low computational complexity despite working at the block-level.

Blind/Referenceless Image Spatial Quality Evaluator

We propose a natural scene statistic based Blind/Referenceless Image Spatial QUality Evaluator (BRISQUE) which extracts the point wise statistics of local normalized luminance signals and measures

No-reference perceptual quality assessment of JPEG compressed images

It is shown that Peak Signal-to-Noise Ratio (PSNR), which requires the reference images, is a poor indicator of subjective quality and tuning an NR measurement model towards PSNR is not an appropriate approach in designing NR quality metrics.

Systematic evaluation of iterative deep neural networks for fast parallel MRI reconstruction with sensitivity‐weighted coil combination

To systematically investigate the influence of various data consistency layers and regularization networks with respect to variations in the training and test data domain, for sensitivity‐encoded

Accelerated MRI With Un-Trained Neural Networks

This work proposes a highly optimized un-trained recovery approach based on a variation of the Deep Decoder and shows that it significantly outperforms otherun-trained methods, in particular sparsity-based classical compressed sensing methods and naive applications of un- trained neural networks in terms of reconstruction performance.

Can Signal-to-Noise Ratio Perform as a Baseline Indicator for Medical Image Quality Assessment

Whether off-the-shelf NIQA models can predict MR image quality is considered by using SNR values as the reference scores, and the correlation betweenSNR values and NIQE results is evaluated.

Perceptual Losses for Real-Time Style Transfer and Super-Resolution

This work considers image transformation problems, and proposes the use of perceptual loss functions for training feed-forward networks for image transformation tasks, and shows results on image style transfer, where aFeed-forward network is trained to solve the optimization problem proposed by Gatys et al. in real-time.