Adaptive intelligence aims at empowering machine learning techniques with the additional use of domain knowledge. In this work, we present the application of adaptive intelligence to accelerate MR acquisition. Starting from undersampled k-space data, an iterative learning-based reconstruction scheme inspired by compressed sensing theory is used to reconstruct the images. We developed a novel deep neural network to refine and correct prior reconstruction assumptions given the training data. The network was trained and tested on a knee MRI dataset from the 2019 fastMRI challenge organized by Facebook AI Research and NYU Langone Health. All submissions to the challenge were initially ranked based on similarity with a known groundtruth, after which the top 4 submissions were evaluated radiologically. Our method was evaluated by the fastMRI organizers on an independent challenge dataset. It ranked #1, shared #1, and #3 on respectively the 8x accelerated multi-coil, the 4x multi-coil, and the 4x single-coil tracks. This demonstrates the superior performance and wide applicability of the method.

BACKGROUND AND PURPOSE:Compressed sensing-sensitivity encoding is a promising MR imaging acceleration technique. This study compares the image quality of compressed sensing-sensitivity encoding accelerated imaging with conventional MR imaging sequences. MATERIALS AND METHODS:Patients with known, treated, or suspected brain tumors underwent compressed sensing-sensitivity encoding accelerated 3D T1-echo-spoiled gradient echo or 3D T2-FLAIR sequences in addition to the corresponding conventional acquisition as part of their clinical brain MR imaging. Two neuroradiologists blinded to sequence and patient information independently evaluated both the accelerated and corresponding conventional acquisitions. The sequences were evaluated on 4-or 5-point Likert scales for overall image quality, SNR, extent/severity of artifacts, and gray-white junction and lesion boundary sharpness. SNR and contrast-to-noise ratio values were compared. RESULTS:Sixty-six patients were included in the study. For T1-echo-spoiled gradient echo, image quality in all 5 metrics was slightly better for compressed sensing-sensitivity encoding than conventional images on average, though it was not statistically significant, and the lower bounds of the 95% confidence intervals indicated that compressed sensing-sensitivity encoding image quality was within 10% of conventional imaging. For T2-FLAIR, image quality of the compressed sensing-sensitivity encoding images was within 10% of the conventional images on average for 3 of 5 metrics. The compressed sensing-sensitivity encoding images had somewhat more artifacts (P ϭ .068) and less gray-white matter sharpness (P ϭ .36) than the conventional images, though neither difference was significant. There was no significant difference in the SNR and contrast-to-noise ratio. There was 25% and 35% scan-time reduction with compressed sensing-sensitivity encoding for FLAIR and echo-spoiled gradient echo sequences, respectively. CONCLUSIONS:Compressed sensing-sensitivity encoding accelerated 3D T1-echo-spoiled gradient echo and T2-FLAIR sequences of the brain show image quality similar to that of standard acquisitions with reduced scan time. Compressed sensing-sensitivity encoding may reduce scan time without sacrificing image quality. ABBREVIATIONS:CNR ϭ contrast-to-noise ratio; CS ϭ compressed sensing; SENSE ϭ sensitivity encoding; SPGR ϭ echo-spoiled gradient echo

Abstract-The problem of output optimization within a specified input space of neural networks (NNs) with fixed weights is discussed in this paper. The problem is (highly) nonlinear when nonlinear activation functions are used. This global optimization problem is encountered in the reinforcement learning (RL) community. Interval analysis is applied to guarantee that all solutions are found to any degree of accuracy with guaranteed bounds. The major drawbacks of interval analysis, i.e., dependency effect and high-computational load, are both present for the problem of NN output optimization. Taylor models (TMs) are introduced to reduce these drawbacks. They have excellent convergence properties for small intervals. However, the dependency effect still remains and is even made worse when evaluating large input do-

No abstract

No abstract