Improving phase‐based conductivity reconstruction by means of deep learning–based denoising of  phase data for 3T MRI

Jung, Kyu Jin; Mandija, Stefano; Kim, Jun Hyeong; Ryu, Kanghyun; Jung, Soozy; Cui, Chuanjiang; Kim, Soo Yeon; Park, Mina; Berg, Cornelis A.T. van den; Kim, Dong Hyun

doi:10.1002/mrm.28826

Cited by 12 publications

(13 citation statements)

References 61 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although the proposed method is expected to be more robust to such factors compared to the conventional phase‐based EPT reconstruction methods (Lee et al, 2021), the experiment results show that noise still impairs the estimation accuracy (Figure 5), and changes in the phase distribution are reflected in the conductivity estimations for the in‐vivo dataset (Figures 6 and 8). Thus, it would be worthwhile to investigate the combination of various artifact reduction or denoising algorithms (Cui et al, 2022; Jung, Mandija, et al, 2021; Michel et al, 2014) with the proposed method to improve the estimation performance for datasets affected by artifacts or high noise.…”

Section: Discussionmentioning

confidence: 99%

Data‐driven electrical conductivity brain imaging using 3 T MRI

Jung

Mandija

Cui

et al. 2023

Human Brain Mapping

Self Cite

View full text Add to dashboard Cite

Magnetic resonance electrical properties tomography (MR‐EPT) is a non‐invasive measurement technique that derives the electrical properties (EPs, e.g., conductivity or permittivity) of tissues in the radiofrequency range (64 MHz for 1.5 T and 128 MHz for 3 T MR systems). Clinical studies have shown the potential of tissue conductivity as a biomarker. To date, model‐based conductivity reconstructions rely on numerical assumptions and approximations, leading to inaccuracies in the reconstructed maps. To address such limitations, we propose an artificial neural network (ANN)‐based non‐linear conductivity estimator trained on simulated data for conductivity brain imaging. Network training was performed on 201 synthesized T2‐weighted spin‐echo (SE) data obtained from the finite‐difference time‐domain (FDTD) electromagnetic (EM) simulation. The dataset was composed of an approximated T2‐w SE magnitude and transceive phase information. The proposed method was tested three in‐silico and in‐vivo on two volunteers and three patients' data. For comparison purposes, various conventional phase‐based EPT reconstruction methods were used that ignore magnitude information, such as Savitzky–Golay kernel combined with Gaussian filter (S‐G Kernel), phase‐based convection‐reaction EPT (cr‐EPT), magnitude‐weighted polynomial‐fitting phase‐based EPT (Poly‐Fit), and integral‐based phase‐based EPT (Integral‐based). From the in‐silico experiments, quantitative analysis showed that the proposed method provides more accurate and improved quality (e.g., high structural preservation) conductivity maps compared to conventional reconstruction methods. Representatively, in the healthy brain in‐silico phantom experiment, the proposed method yielded mean conductivity values of 1.97 ± 0.20 S/m for CSF, 0.33 ± 0.04 S/m for WM, and 0.52 ± 0.08 S/m for GM, which were closer to the ground‐truth conductivity (2.00, 0.30, 0.50 S/m) than the integral‐based method (2.56 ± 2.31, 0.39 ± 0.12, 0.68 ± 0.33 S/m). In‐vivo ANN‐based conductivity reconstructions were also of improved quality compared to conventional reconstructions and demonstrated network generalizability and robustness to in‐vivo data and pathologies. The reported in‐vivo brain conductivity values were in agreement with literatures. In addition, the proposed method was observed for various SNR levels (SNR levels = 10, 20, 40, and 58) and repeatability conditions (the eight acquisitions with the number of signal averages = 1). The preliminary investigations on brain tumor patient datasets suggest that the network trained on simulated dataset can generalize to unforeseen in‐vivo pathologies, thus demonstrating its potential for clinical applications.

show abstract

Section: Discussionmentioning

confidence: 99%

Data‐driven electrical conductivity brain imaging using 3 T MRI

Jung

Mandija

Cui

et al. 2023

Human Brain Mapping

Self Cite

View full text Add to dashboard Cite

show abstract

“…The training was done without any prior knowledge of physics embedded in analytic models. For denoising, a noise filtering method for the

-fields measurements was proposed, where a NN black-box model was learned from numerically simulated samples for filtering the

[ 22 ]. Although it shows improved accuracy for numerically simulated samples, it lacks generalization or robustness for its clinical uses [ 23 ].…”

Section: Introductionmentioning

confidence: 99%

“…As a matter of fact, big clinical datasets for training the NNs for MREPT are not available. Therefore, the datasets for training have been constructed from numerically simulated samples [ 21 , 22 , 24 , 27 ]. While simulated datasets are appropriate for many applications, there are limitations of such datasets in the medical image reconstruction field [ 28 ].…”

Section: Introductionmentioning

confidence: 99%

Physics Informed Neural Networks (PINN) for Low Snr Magnetic Resonance Electrical Properties Tomography (MREPT)

et al. 2022

View full text Add to dashboard Cite

Electrical properties (EPs) of tissues facilitate early detection of cancerous tissues. Magnetic resonance electrical properties tomography (MREPT) is a technique to non-invasively probe the EPs of tissues from MRI measurements. Most MREPT methods rely on numerical differentiation (ND) to solve partial differential Equations (PDEs) to reconstruct the EPs. However, they are not practical for clinical data because ND is noise sensitive and the MRI measurements for MREPT are noisy in nature. Recently, Physics informed neural networks (PINNs) have been introduced to solve PDEs by substituting ND with automatic differentiation (AD). To the best of our knowledge, it has not been applied to MREPT due to the challenges in using PINN on MREPT as (i) a PINN requires part of ground-truth EPs as collocation points to optimize the network’s AD, (ii) the noisy input data disrupts the optimization of PINNs despite the noise-filtering nature of NNs and additional denoising processes. In this work, we propose a PINN-MREPT model based on a canonical analytic MREPT model. A reference padding layer with known EPs was added to surround the region of interest for providing additive collocation points. Moreover, an optimizable diffusion coefficient was embedded in the analytic MREPT model used in the PINN-MREPT. The noise robustness of the proposed PINN-MREPT for single-sample reconstruction was tested by using numerical phantoms of human brain with extra tumor-like tissues at different noise levels. The results of numerical experiments show that PINN-MREPT outperforms two typical numerical MREPT methods in terms of reconstruction accuracy, sensitivity to the extra tissues, and the correlations of line profiles in the regions of interest. The advantage of the PINN-MREPT is shown by the results of an experiment on phantom measurement, too. Moreover, it is found that the diffusion term plays an important role to achieve a noise-robust PINN-MREPT. This is an important step moving forward to a clinical application of MREPT.

show abstract

“…In most deep learning based coherent noise suppression methods, the training data pairs of noisy and noise-free phase images are first constructed using the noise model. The training data pairs are utilized to train a neural network capable mapping both the noisy and noise-free phase images [16][17][18][19][20]. However, these methods require paired data to train the neural network, which is difficult in practical holographic applications.…”

Section: Introductionmentioning

confidence: 99%

Coherent noise suppression in digital holographic microscopy based on label-free deep learning

Tang

Zhang

et al. 2022

Front. Phys.

View full text Add to dashboard Cite

Deep learning techniques can be introduced into the digital holography to suppress the coherent noise. It is often necessary to first make a dataset of noisy and noise-free phase images to train the network. However, noise-free images are often difficult to obtain in practical holographic applications. Here we propose a label-free training algorithms based on self-supervised learning. A dilated blind spot network is built to learn from the real noisy phase images and a noise level function network to estimate a noise level function. Then they are trained together via maximizing the constrained negative log-likelihood and Bayes’ rule to generate a denoising phase image. The experimental results demonstrate that our method outperforms standard smoothing algorithms in accurately reconstructing the true phase image in digital holographic microscopy.

show abstract

Improving phase‐based conductivity reconstruction by means of deep learning–based denoising of phase data for 3T MRI

Cited by 12 publications

References 61 publications

Data‐driven electrical conductivity brain imaging using 3 T MRI

Data‐driven electrical conductivity brain imaging using 3 T MRI

Physics Informed Neural Networks (PINN) for Low Snr Magnetic Resonance Electrical Properties Tomography (MREPT)

Coherent noise suppression in digital holographic microscopy based on label-free deep learning

Contact Info

Product

Resources

About