skip to main content
US FlagAn official website of the United States government
dot gov icon
Official websites use .gov
A .gov website belongs to an official government organization in the United States.
https lock icon
Secure .gov websites use HTTPS
A lock ( lock ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.


Title: Revisiting l1-wavelet compressed-sensing MRI in the era of deep learning
Following their success in numerous imaging and computer vision applications, deep-learning (DL) techniques have emerged as one of the most prominent strategies for accelerated MRI reconstruction. These methods have been shown to outperform conventional regularized methods based on compressed sensing (CS). However, in most comparisons, CS is implemented with two or three hand-tuned parameters, while DL methods enjoy a plethora of advanced data science tools. In this work, we revisit 1 -wavelet CS reconstruction using these modern tools. Using ideas such as algorithm unrolling and advanced optimization methods over large databases that DL algorithms utilize, along with conventional insights from wavelet representations and CS theory, we show that 1 -wavelet CS can be fine-tuned to a level close to DL reconstruction for accelerated MRI. The optimized 1 -wavelet CS method uses only 128 parameters compared to >500,000 for DL, employs a convex reconstruction at inference time, and performs within <1% of a DL approach that has been used in multiple studies in terms of quantitative quality metrics.  more » « less
Award ID(s):
1651825
PAR ID:
10469961
Author(s) / Creator(s):
; ; ; ; ;
Publisher / Repository:
Proc Natl Acad Sci USA
Date Published:
Journal Name:
Proceedings of the National Academy of Sciences
Volume:
119
Issue:
33
ISSN:
0027-8424
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract Electroencephalograms (EEG) are invaluable for treating neurological disorders, however, mapping EEG electrode readings to brain activity requires solving a challenging inverse problem. For time series data, the use of 1 regularization quickly becomes intractable for many solvers, and, despite the reconstruction advantages of 1 regularization, 2 -based approaches such as standardized low-resolution brain electromagnetic tomographysLORETAare used in practice. In this work, we formulate EEG source localization as a graphical generalized elastic net inverse problem and present avariable projectedaugmented Lagrangian algorithm (VPAL) suitable for fast EEG source localization. We prove convergence of this solver for a broad class of separable convex, potentially non-smooth functions subject to linear constraints. Leveraging the efficiency of the proposedVPALalgorithm, we introduce a windowed variation,VPAL W , that computes time dynamics in sequence suitable for real-time reconstruction. Our proposed methods are compared to state-of-the-art approaches includingsLORETAand other methods for 1 -regularized inverse problems. 
    more » « less
  2. We combine synchrotron-based infrared absorption and Raman scattering spectroscopies with diamond anvil cell techniques and first-principles calculations to explore the properties of hafnia under compression. We find that pressure drives HfO 2 :7%Y from the mixed monoclinic ( P 2 1 / c ) + antipolar orthorhombic ( Pbca ) phase to pure antipolar orthorhombic ( Pbca ) phase at approximately 6.3 GPa. This transformation is irreversible, meaning that upon release, the material is kinetically trapped in the Pbca metastable state at 300 K. Compression also drives polar orthorhombic ( P c a 2 1 ) hafnia into the tetragonal ( P 4 2 / n m c ) phase, although the latter is not metastable upon release. These results are unified by an analysis of the energy landscape. The fact that pressure allows us to stabilize targeted metastable structures with less Y stabilizer is important to preserving the flat phonon band physics of pure HfO 2
    more » « less
  3. MXenes have demonstrated potential for various applications owing to their tunable surface chemistry and metallic conductivity. However, high temperatures can accelerate MXene film oxidation in air. Understanding the mechanisms of MXene oxidation at elevated temperatures, which is still limited, is critical in improving their thermal stability for high-temperature applications. Here, we demonstrate that Ti 3 C 2 T x MXene monoflakes have exceptional thermal stability at temperatures up to 600 ° C in air, while multiflakes readily oxidize in air at 300 ° C. Density functional theory calculations indicate that confined water between Ti 3 C 2 T x flakes has higher removal energy than surface water and can thus persist to higher temperatures, leading to oxidation. We demonstrate that the amount of confined water correlates with the degree of oxidation in stacked flakes. Confined water can be fully removed by vacuum annealing Ti 3 C 2 T x films at 600 ° C, resulting in substantial stability improvement in multiflake films (can withstand 600 ° C in air). These findings provide fundamental insights into the kinetics of confined water and its role in Ti 3 C 2 T x oxidation. This work enables the use of stable monoflake MXenes in high-temperature applications and provides guidelines for proper vacuum annealing of multiflake films to enhance their stability. 
    more » « less
  4. Earthquakes on the Salt Lake City Segment of the Wasatch fault (WFSLC) represent the most significant seismic hazard to the Salt Lake Valley, populated by 1 million+ people. The 2020 Magna, UT, earthquake, which likely occurred on the WFSLC, generated peak ground accelerations (PGAs) as large as 0.55 gin the Salt Lake Valley. Here, we present three-dimensional (3D) physics-based wave propagation simulations of the Magna earthquake sequence in the Wasatch Front Community Velocity Model (WFCVM) up to 10 Hz to better constrain both linear and nonlinear parameters in the soils of the Salt Lake Valley. We first calibrate the WFCVM via linear simulations of a M w 4.59 Magna aftershock, obtaining the best fit between the recordings and synthetics, including a statistical distribution of small-scale heterogeneities with 10% standard deviation and Q S = 0 . 05 V S for frequencies < 1  Hz and Q S = 0 . 05 V S f 0 . 4 for frequencies > 1  Hz ( V s in m/s). Spectral ratios from our simulations of the 2020 Magna mainshock using a finite-fault source model generally overestimate those for the recordings in the linear regime at higher frequencies, in particular at stations with the largest PGAs, suggesting the presence of nonlinear soil effects. Using a fully hysteretic multi-yield-surface 3D nonlinear modeling approach, we find that damping from the reference strain–depth relations proposed by Darendeli significantly reduces the bias in terms of spectral amplification ratios at stations with the shortest epicentral distances. We find an optimal fit between the recordings and nonlinear synthetics for reference strains at about 2 standard deviations below Darendeli’s relations, with reduction of the spectral amplification bias by more than a factor of two. Our findings suggest significant nonlinear soil effects in the Salt Lake Valley and provide a basis for improved seismic hazard analysis of the greater Salt Lake City region. 
    more » « less
  5. Assessment of the global budget of the greenhouse gas nitrous oxide ( N 2 O) is limited by poor knowledge of the oceanic N 2 O flux to the atmosphere, of which the magnitude, spatial distribution, and temporal variability remain highly uncertain. Here, we reconstruct climatological N 2 O emissions from the ocean by training a supervised learning algorithm with over 158,000 N 2 O measurements from the surface ocean—the largest synthesis to date. The reconstruction captures observed latitudinal gradients and coastal hot spots of N 2 O flux and reveals a vigorous global seasonal cycle. We estimate an annual mean N 2 O flux of 4.2 ± 1.0 Tg N y 1 , 64% of which occurs in the tropics, and 20% in coastal upwelling systems that occupy less than 3% of the ocean area. This N 2 O flux ranges from a low of 3.3 ± 1.3 Tg N y 1 in the boreal spring to a high of 5.5 ± 2.0 Tg N y 1 in the boreal summer. Much of the seasonal variations in global N 2 O emissions can be traced to seasonal upwelling in the tropical ocean and winter mixing in the Southern Ocean. The dominant contribution to seasonality by productive, low-oxygen tropical upwelling systems (>75%) suggests a sensitivity of the global N 2 O flux to El Niño–Southern Oscillation and anthropogenic stratification of the low latitude ocean. This ocean flux estimate is consistent with the range adopted by the Intergovernmental Panel on Climate Change, but reduces its uncertainty by more than fivefold, enabling more precise determination of other terms in the atmospheric N 2 O budget. 
    more » « less