Deep learning using a biophysical model for robust and accelerated reconstruction of quantitative, artifact‐free and denoised  images

Torop, Max; Kothapalli, Satya V. V. N.; Sun, Yu; Liu, Jiaming; Kahali, Sayan; Yablonskiy, Dmitriy A.; Kamilov, Ulugbek S.

doi:10.1002/mrm.28344

Cited by 22 publications

(42 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our approach is based on the qGRE method utilizing a multi-gradient-echo MRI sequence available from most MRI manufacturers and required only about 6 minutes of MRI scanning time. Data analysis can be significantly accelerated with the aid of deep learning method [73], thus opening opportunity for broad research and clinical applications. Combining in vivo qGRE biomarkers of neuronal injury with the current in vivo PET and CSF biomarkers would allow for better understanding and detection of the early pathology in Alzheimer disease and other dementias.…”

Section: Discussionmentioning

confidence: 99%

Quantitative Gradient Echo MRI Identifies Dark Matter as a New Imaging Biomarker of Neurodegeneration that Precedes Tissue Atrophy in Early Alzheimer Disease

Benzinger

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Background: Currently, brain tissue atrophy serves as in vivo MRI biomarker of neurodegeneration in Alzheimer Disease (AD). However, postmortem histopathological studies show that neuronal loss in AD exceeds volumetric loss of tissue and that loss of memory in AD begins when neurons and synapses are lost. Therefore, in vivo detection of neuronal loss prior to detectable atrophy in MRI is essential for early AD diagnosis. Objective: To apply a recently developed quantitative Gradient Recalled Echo (qGRE) MRI technique for in vivo evaluation of neuronal loss in human hippocampus. Methods: Seventy participants were recruited from the Knight Alzheimer Disease Research Center, representing three groups: Healthy controls [Clinical Dementia Rating® (CDR®)=0, amyloid β ; (Aβ ;)-negative), n=34]; Preclinical AD (CDR=0, Aβ ;-positive, n=19); and mild AD (CDR=0.5 or 1, Aβ ;-positive, n=17). Results: In hippocampal tissue, qGRE identified two types of regions: one, practically devoid of neurons, we designate as "Dark Matter", the other, with relatively preserved neurons - "Viable Tissue". Data showed a greater loss of neurons than defined by atrophy in the mild AD group compared with the healthy control group - neuronal loss ranged between 31% and 43% while volume loss ranged only between 10% and 19%. The concept of Dark Matter was confirmed with histopathological study of one participant who underwent in vivo qGRE 14 months prior to expiration. Conclusion: in vivo qGRE method identifies neuronal loss that is associated with impaired AD-related cognition but is not recognized by MRI measurements of tissue atrophy, therefore providing new biomarkers for early AD detection.

show abstract

Section: Discussionmentioning

confidence: 99%

Quantitative Gradient Echo MRI Identifies Dark Matter as a New Imaging Biomarker of Neurodegeneration that Precedes Tissue Atrophy in Early Alzheimer Disease

Benzinger

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…In this context, self-supervised learning approaches are more suitable for dealing with limited data sets, such as the one presented in this work. Self-supervised learning in biomedical imaging has been implemented in several instances, among which screening of 2-dimensional chest x-ray images [29], in the evaluation of cardiac time-series data [30], in tissue segmentation of brain lesions [31], in segmentation of the renal dynamic contrast-enhanced MRI [32], in robust and accelerated reconstruction of quantitative and B 0 -inhomogeneity-corrected R * 2 maps from multi-gradient recalled echo MRI data [33] and in quality enhancement of compressed sensing MRI of vessel wall [34].…”

Section: Introductionmentioning

confidence: 99%

MRI-based classification of neuropsychiatric systemic lupus erythematosus patients with self-supervised contrastive learning

Inglese

Kim

et al. 2021

Preprint

View full text Add to dashboard Cite

Introduction/PurposeSystemic lupus erythematosus (SLE) is a chronic auto-immune disease with a broad spectrum of clinical presentations, including heterogeneous and uncommon neuropsychiatric (NP) syndromes. Accurate diagnosis of neuropsychiatric SLE (NPSLE) is challenging due to lack of clinically useful biomarkers. Despite structural brain abnormalities on MRI in NPSLE being a common finding, a robust link between structural abnormalities and NPSLE has not been established, thus their contribution to the distinction between NPSLE patients and patients in which the NP symptoms are not primarily attributed to SLE is limited. Self-supervised contrastive learning algorithms do not require labels, and have been shown to be useful in classification tasks in rare diseases with limited number of datasets. The aim of our study was to apply self-supervised contrastive learning on T1-weighted images acquired from a well-defined cohort of SLE patients to distinguish between SLE patients with NP symptoms due to the disease (NPSLE) or and SLE patients with similar symptoms due to other causes (non-NPSLE).Subjects and Methods163 patients were included. We used 3T MRI T1-weighted images registered to the MNI152 template. The training set comprised 68 non-NPSLE and 34 NPSLE patients. During the training procedure, we applied random geometric transformations (cropping, left-right flipping and rotations) between iterations to enrich our data sets. Our ML pipeline consisted of convolutional base encoder and linear projector. To test the classification task, the projector was removed and one linear layer was measured. We trained the encoder and projector with the Normalized Temperature-scaled Cross Entropy Loss (NT-xent) loss function. We performed a Monte Carlo validation that consisted of 6 repeated random sub-samplings each using a random selection of a small group of samples from each group.ResultsIn the 6 trials described above, between 79% and 83% of the patients were correctly classified as NPSLE or non-NPSLE. For a qualitative evaluation of spatial distribution of the common features found in the NPSLE population, Gradient-weighted Class Activation Maps (Grad-CAM) were examined voxel-wise. Thresholded Grad-CAM maps show areas of common features identified for the NPSLE cohort, with no such communality found for the non-NPSLE group.Discussion/conclusionThe self-supervised contrastive learning model was effective in capturing diagnostic brain MRI features from a limited but well-defined cohort of SLE patients with NP symptoms. The interpretation of the Grad-CAM results is not straightforward, but points to involvement of the lateral and third ventricles, periventricular white matter and basal cisterns. We believe that the common features found in the NPSLE population in this study indicate a combination of tissue loss, local atrophy and to some extent that of periventricular white matter lesions, which are commonly found in NPSLE patients and appear hypointense on T1-weighted images.

show abstract

“…Recently we have demonstrated (Torop et al, 2020) that using deep learning approach allows reconstruction of combined R2* maps in a matter of seconds with improved image quality and reduced noise effects. However, separate generating quantitative maps of R2t* and 2′…”

Section: Introductionmentioning

confidence: 99%

“…Wang, Sukstanskii, & Yablonskiy, 2013;Zhao et al, 2016) not always available from experimental data. From this perspective, using deep neural network instead of ANN provides clear advantage in dealing with noisy data (Torop et al, 2020) and also reducing computation time. Moreover, we have developed a deep learning framework (Xu et al, 2021) which performs motion correction on complex qGRE images and enables reconstruction of high-quality motion-free quantitative R2* maps.…”

Section: Introductionmentioning

confidence: 99%

“…Our method is based on Convolutional Neural Network (CNN) which is trained on qGRE MR images reflecting signal decay. The network architecture is similar to our recently published RoAR method (Torop et al, 2020). However, the ground truth maps were generated by fitting the biophysical model (Ulrich & Yablonskiy, 2016) to experimental GRE data with multiple gradient echoes using nonlinear least square algorithm.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Deep-Learning-Based Accelerated and Noise-Suppressed Estimation (DANSE) of quantitative Gradient Recalled Echo (qGRE) MRI metrics associated with Human Brain Neuronal Structure and Hemodynamic Properties

Kahali

Kothapalli

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Purpose: To introduce a Deep-Learning-Based Accelerated and Noise-Suppressed Estimation (DANSE) method for reconstructing quantitative maps of biological tissue cellular-specific, and hemodynamic-specific, from Gradient-Recalled-Echo (GRE) MRI data with multiple gradient-recalled echoes. Methods: DANSE method adapts supervised learning paradigm to train a convolutional neural network for robust estimation of and maps free from the adverse effects of macroscopic (B0) magnetic field inhomogeneities directly from the GRE magnitude images without utilizing phase images. The corresponding ground-truth maps were generated by means of a voxel-by-voxel fitting of a previously-developed biophysical quantitative GRE (qGRE) model accounting for tissue, hemodynamic and -inhomogeneities contributions to GRE signal with multiple gradient echoes using nonlinear least square (NLLS) algorithm. Results: We show that the DANSE model efficiently estimates the aforementioned brain maps and preserves all features of NLLS approach with significant improvements including noise-suppression and computation speed (from many hours to seconds). The noise-suppression feature of DANSE is especially prominent for data with SNR characteristic for typical GRE data (SNR~50), where DANSE-generated and maps had three times smaller errors than that of NLLS method. Conclusions: DANSE method enables fast reconstruction of magnetic-field-inhomogeneity-free and noise-suppressed quantitative qGRE brain maps. DANSE method does not require any information about field inhomogeneities during application. It exploits spatial patterns in the qGRE MRI data and previously-gained knowledge from the biophysical model, thus producing clean brain maps even in the environments with high noise levels. These features along with fast computational speed can lead to broad qGRE clinical and research applications.

show abstract

Deep learning using a biophysical model for robust and accelerated reconstruction of quantitative, artifact‐free and denoised images

Cited by 22 publications

References 32 publications

Quantitative Gradient Echo MRI Identifies Dark Matter as a New Imaging Biomarker of Neurodegeneration that Precedes Tissue Atrophy in Early Alzheimer Disease

Quantitative Gradient Echo MRI Identifies Dark Matter as a New Imaging Biomarker of Neurodegeneration that Precedes Tissue Atrophy in Early Alzheimer Disease

MRI-based classification of neuropsychiatric systemic lupus erythematosus patients with self-supervised contrastive learning

Deep-Learning-Based Accelerated and Noise-Suppressed Estimation (DANSE) of quantitative Gradient Recalled Echo (qGRE) MRI metrics associated with Human Brain Neuronal Structure and Hemodynamic Properties

Contact Info

Product

Resources

About