Approaches to magnetic resonance (MR) image acquisition and computational modelling often exist in silo. The application of computational modelling to paediatric data is in its infancy. This collaborative study addresses this gap for multiple organ systems. State-of-the art MR imaging sequences were selected and refined to provide imaging data for the creation of subject-specific geometric models. The MR protocol and modelling outputs will underpin a longitudinal paediatric study set in Aotearoa New Zealand, generating normative data in healthy children, and identifying early biomarkers of pathological processes. This will enable the development of new personalised, predictive, and preventative models of care.

When analyzing multi-site diffusion MRI (dMRI) data, metrics should be harmonized to remove the site effect. In this study, we applied the combined association test (ComBat), which uses regression of covariates with an empirical Bayes framework, for diffusion metrics based on diffusion tensor imaging (DTI) and neurite orientation dispersion and density imaging (NODDI). The results showed that ComBat can harmonize site-related effects in DTI and NODDI metrics based on multi-site dMRI while preserving subject biological information, such as sex differences and correlation with age. Thus, ComBat could be applied in large multi-site studies to identify subtle white matter changes.

¹H Magnetic Resonance Spectroscopy (MRS) suffers low Signal-Noise Ratio (SNR) due to low concentrations of metabolites. To improve the SNR, the current mainstream is to do Signal Averaging with  repeated samplings but it is time-consuming. Therefore, we designed a novel denoising ReLSTM-Net to learn the mapping from the low SNR MRS  to the high SNR one  in the time-domain by a few in vivo measured data. Denoised spectra by the proposed method has higher accuracy and reliability in quantifying metabolites Glx, tCho and mI, compared with the state-of-art Low-Rank method.

Computer aided diagnosis (CAD) is widely considered an important application of deep learning in healthcare. However, brain data with lesion location labels are rare in MRI domain. Recently, Microsoft Research has released a dataset of clinical pathology annotations based on the raw images from fastMRI and named it fastMRI+. Here, fastMRI+ brain dataset was analyzed and used to train deep learning-based models for lesion detection. Some well-known object detection architectures such as YOLOv3, Faster-RCNN and YOLOX  were compared. Overall, this abstract established a baseline with improvement suggestions for future studies.

In this work noise is assumed to be a random vector and the method of unbiased predictive risk estimator (UPRE) is used to select suitable data/regularization parameters to solve the local phase to susceptibility deconvolution problem of quantitative susceptibility mapping (QSM). The proposed algorithm is tested on the simulated multi-echo data provided at the 2019 QSM Reconstruction Challenge. This work is a further development of the algorithm presented at the ISMRM Meeting 2021 and its purpose is to show that the method of UPRE can be applied advantageously to a shearlet /TGV based susceptibility reconstruction. 

Abnormal placental development is postulated as one of the leading causes of fetal growth restriction (FGR). Advances in MRI technology enable non-invasive measurement of fetal oxygen saturation, but not all have yet been validated. Due to the invasiveness of tests required, validation in human subjects is not possible. Preclinical models such as pregnant sheep allow invasive methods to validate MRI measurements. Here we show that multi-compartment modelling of non-invasive placental MRI can be used to estimate the oxygen saturation of feto-placental oxygenation in normal sheep pregnancy and pregnancies affected by early gestation onstet FGR.

In this abstract, we describe our findings from the phenotypic association analyses (univariate Pearson correlations) between 17,485 non-imaging phenotypes and QSM (and T2*) measures using data from 35,885 subjects in UK Biobank. In total, we identify statistically-significant associations of 251 phenotypes with QSM IDPs. Here we describe example associations in blood assays, health outcomes, food and drink, and alcohol consumption categories in detail. Our results demonstrate that QSM and T2* contribute complementary information. This new QSM resource provides an opportunity to investigate susceptibility contrast in previously unexplored territory, which might lead to advances in application of QSM in neuroscience.

The subthalamic structures, like the zona incerta (ZI), fields of forel (FF), subthalamic nucleus (STN), and their surrounding structures, are poorly characterized regarding anatomical imaging. Using quantitative susceptibility mapping (QSM) at ultra-high-field, iron, and myelin mapping may allow an advanced characterization. Therefore, QSM data were acquired at 9.4 Tesla in 21 subjects to characterize the para- and diamagnetic properties of the ZI, FF, and their adjacent structures. In contrast to the FF, we found a higher contribution of diamagnetic and paramagnetic sources, i.e.,  iron, myelin, and calcium, in the STN and ZI. 

Slice-specific z-shimming reduces signal losses caused by through-slice dephasing in spinal cord fMRI acquisitions. Typically, the optimum values are determined from a reference acquisition covering a range of z-shim values by a user who identifies the setting with the least signal loss. However, this procedure requires some experience, is time-consuming, and user-dependent. Here, an automated calculation of the optimum values has been developed, implemented on the MR system, and evaluated in phantoms and in vivo. Without compromising the image quality, the approach is faster and user-independent and, thus, may help to facilitate the application of z-shimming in spinal cord fMRI.

This study proposes a rapid and robust ASL-based time-resolved MRA technique with high spatiotemporal resolution termed Dual-Subspace MRA (DS-MRA), which employs subspace modeling from both temporal and control/label dimensions as sparsity constraints to improve the robustness of image reconstruction with under-sampled golden-angle radial dynamic MRA. The performance of DS-MRA was compared with conventional iGRASP and 5-dimensional GRASP. Our preliminary data suggests that DS-MRA outperforms conventional GRASP reconstructions with less residual streaking artifacts and noise, especially at higher acceleration rates.

Quantitative imaging has been very useful in neuroscientific and clinical applications, including glioma, tumor diagnosis and prognosis, brain maturation, and Alzheimer's disease. EPI is a powerful tool for quantitative imaging owing to its extremely fast acquisition. This work aims to develop a distortion-free, blip-up/down acquisition (BUDA) 3D-EPI with controlled aliasing in parallel imaging (CAIPI) sampling and joint Hankel low-rank image reconstruction for fast and robust multi-contrast high-resolution whole-brain imaging. The developed technique could generate distortion-free high-resolution whole-brain T2* mapping and quantitative susceptibility mapping in 47s at 1.1×1.1×1 mm³ resolution.

Inter-shot phase correction is a critical step in multi-band multi-shot diffusion MRI. The phase correction can be accomplished by first estimating the explicit phase map and then inputting it into the diffusion signal formulation model to recover the diffusion images. Alternatively, the phase information can be used in an indirect manner to determine structured low rank constraints in k-space. The two methods differ in terms of reconstruction accuracy and efficiency. In this study, we propose a new way to combine the two approaches for improved image quality, termed “Joint Usage of structured Low-rank constraints and Explicit Phase mapping” (JULEP). 

It is shown that spectral overlap can cause an apparent correlation between metabolites regardless of underlying biological correlations or the lack thereof. Because MRS signals are often used to correlate with other MRS signals or clinical measures a theoretical framework is developed to estimate correlations originating from spectral overlap at long echo time. Monte Carlos simulations were also performed to quantify the correlations. Our results showed that significant correlations may occur even at long TE, when the contribution from macromolecule background becomes negligible. The proposed theoretical framework was proven to be useful for predicting cross-correlation coefficients originating from spectral overlap.    

Efforts have been focused on early detection of osteoarthritis using MRI. Our study reports T1, T1ρ and T2 values for knee cartilage and menisci in normal subjects using 3T MRI, and evaluates associations with age, gender, and BMI. We found that WB-FC had higher T1 and lower T2 values than LWB-FC, and WB-FC had higher T1 values in females. T1 values of WB-FC and LM were associated with age, and T1ρ values of WB-FC were associated with BMI. Our study presents novel data on T1 mapping, reinforcing the utility of MRI as a potential tool for early detection of osteoarthritis.

Recent developments in spatiotemporal MRI techniques enable whole-brain multi-parametric mapping in incredibly short acquisition times through highly-efficient k-space encoding, subspace reconstruction and carefully-designed regularization. However, this comes at the cost of long reconstruction times making such methods difficult to integrate into clinical practice.

This abstract proposes a framework denoted SMILR (pronounced smile-r) to reduce the reconstruction times of subspace methods from multiple hours to a few minutes through machine learning. To evaluate performance, the framework is applied to multi-axis spiral projection MRF (denoted SPI-MRF) where it achieves improved reconstruction over conventional subspace reconstruction with locally low-rank at ~16-20x faster speed.

Deep Learning based reconstruction methods have been vastly explored for accelerating Cartesian-based cardiac cine imaging via using unrolled neural networks. However, for non-Cartesian trajectories such as radial, these networks require substantial modifications (i.e., NUFFT-based data consistency) and requires collecting separate radial-based dataset, which may not be common in the clinics. 

Here, we investigate a method to transfer the radial k-space data to the Cartesian domain using GROG-based interpolation. We show that DL-ESPIRiT trained with Cartesian cine dataset (with pseudo radial-like under sampling pattern) can be generalizable to reconstruct actual accelerated radial cine acquired on a scanner.

We reconstruct mouse optic pathways with ODF-Fingerprinting in a challenging crossing area of the optic chiasm. Our method replaces the commonly used peak finding approach with pattern matching when reconstructing white matter fiber directions from orientation distribution functions. As reference, we perform analogous reconstructions using Radial Diffusion Spectrum Imaging and Generalized Q-space Imaging. To obtain ground truth information, we inject intravitreally a 0.1M solution of manganese chloride into four C57BL/6N mice. The optic pathways reconstructed with deterministic tractography based on directional information calculated with our approach reach higher agreement with the ground truth than the other two methods.

In this work we propose a novel algorithm of denoising accelerated diffusion weighted MRI (dMRI) acquisitions using deep learning and self-supervision. This method effectively enables the prediction of diffusion-weighted images (DWIs), without the need for large amounts of training data with high directional encodings. We demonstrate that accurate diffusion tensor metrics can be obtained with as few as 6 DWIs using only a few training datasets with high directional encodings.

We propose an efficient, densely connected network to synthesize unacquired DW-volumes from other acquired DW-directions and b=0 mm²/s volumes, allowing acceleration of high-angular DWI shell acquisition by skipping some DW-directions altogether. 

 For training, we used a high-quality dataset of 20 HCP subjects with 90 DW-directions per subject at both b=1000 mm²/s and 3000 mm²/s. 40 HCP subjects were used for validation. 30 DW-directions were selected for input to reconstruct the other 60 missing target DW-directions.

Comparison with a linear-interpolation benchmark show improved fidelity of synthesized DW-volumes and FOD maps to a gold standard acquisition, for both b-values.

In this study, we trained a fully automatic lesion segmentation model of venous malformations based on Unet++ structure and fat-saturated T2-weighted images. The results of automatic segmentation of lesions in different sequence directions and locations of the lesions are consistent with the results of manual segmentation by radiologists. The Dice coefficient on the test set reached 0.847. This fully automatic lesion segmentation model can provide support for subsequent automatic diagnosis studies related to venous malformations.

Monte Carlo modeling enables characterization of MR signals in various tissues, and has been applied to liver MR in the presence of fat. However, Monte Carlo modeling requires accurate information about the underlying tissue properties. In this work, we investigate the size and clustering of fat droplets in the liver using stereology and spatial statistics for three human liver biopsy samples with steatosis. Results show that the generalized gamma distribution function can accurately determine the size and location distributions of fat droplets. This may enable analysis of the underlying biophysical mechanisms between fat fraction and R2* from microscopic magnetic sensitivity.

Segmentation and volumetric analysis of the hypothalamus and fornix plays a critical role in improving the understanding of degenerative processes that might impact the function of these structures. We present Open-Source Hypothalamic-ForniX (OSHy-X) atlases and tool for multi-atlas fusion segmentation for 3T and 7T. The atlases are based on 20 manual segmentations, which we demonstrate have high interrater agreement. The versatility of the OSHy-X tool allows segmentation and volumetric analysis for different field strengths and contrasts. We also demonstrate that OSHy-X segmentation has higher Dice overlaps (3T and 7T inputs: p<0005, p<0.005) than a deep-learning segmentation method for the hypothalamus.

Deep-learning-based slice-by-slice noise reduction may not be suitable for low-SNR body DWI that contains insufficient information in the original single slice. Moreover, averaging multiple acquisitions after denoising to avoid this problem is insufficient because it causes blurring owing to a mismatch between acquisitions. Herein, we designed a neural network that utilises 4D convolution to incorporate adjacent slices and multiple acquisitions simultaneously for a slice to achieve adequate denoising. The results support the utility of the proposed method in comparison with the usual slice-by-slice method and averaging.

As grade Ⅱ and Ⅲ gliomas are difficult to distinguish in preoperative, this study attempted to find the best perfusion parameters for identifying grade Ⅱ/Ⅲ glioma by machine learning model. The machine learning model showed robust performance when using the parameters of volume transfer coefficient (K^trans) and mean transit time (MTT) derived from the dynamic contrast-enhanced (DCE) and dynamic susceptibility contrast (DSC) imaging, which indicated that the combination of DCE and DSC perfusion techniques is expected to further improve the differential diagnosis of grade Ⅱ and Ⅲ gliomas.

The current study systematically investigated multiple factors that affect the reliability of the FC location and strength of the DLPFC with a few emotional related deep regions. The group-level voxel-wise FC strength reliability was low to moderate, indicating its little significance for guiding individualized rTMS treatment. In terms of the FC location reliability, the intra-individual distances of the center of gravity (COG) were 3.8-7.3 mm across different conditions, which suggest that the COG of the seed-based FC might be potential stimulation target for individualized precise rTMS treatment of affective brain disorders.

To detect distinctive structural abnormalities of schizophrenia early in magnetic resonance imaging (MRI) data, we propose a 3D Medical Image Global-Regional (3D-MIGR) Transformer with a VGG11BN backbone followed by a global-regional transformer encoder, which outperforms state-of-the-art models. We trained and tested our model on 887 pre-processed structural whole-head (WH) T1W 3D images from 3 datasets with similar acquisition parameters. 3D-MIGR Transformer improves AUROC to 0.990 and demonstrates strong generality. We found that combining volume-level contextual information with patch-level features enhances performance and allows us to identify ventricle areas as the most informative regions in regional schizophrenia likelihood visualization.

Using a simple linear model, we predicted myelin-specific PSR maps from clinically available images. We trained this linear model on various combinations of routine images, including gray-matter standardized T₁- and T₂-weighted images and diffusion maps. The model reasonably predicted PSR values from structural images alone, although model performance was improved by the addition of diffusion parameters. In the future, this model may enable researchers and clinicians to assess myelin status without the need for specialized myelin imaging sequences. In addition, myelin maps could be obtained retrospectively in large imaging databases without myelin specific methods.

Finite Element (FE)-based mechanical models can simulate the brain growth and folding process. But they are time consuming due to the large number of nodes in a real human brain and the reverse process to the initial smooth brain surfaces is difficult because it is not invertible problem. Here, we demonstrate a proof-of-concept that deep-learning neural networks (DNN) can learn the growth and folding process of human brain in forward and reverse directions and can predict/retrieve the developed/primary folding patterns in a very fast speed.

Applications such as PET/MR and MR-only Radiotherapy Planning need the capability to derive a CT-like image from MRI inputs to enable accurate attenuation correction and dose estimation. More recently, transformer models have been proposed for computer vision applications. Models in the transformer family discard traditional convolution-based network structures and emphasize the importance of non-local information yielding potentially more realistic outputs. To evaluate the performance of SwinIR, TransUet, and Unet. After comparing results visually and quantitatively, the SwinIR models and TransUnet models appear to provide higher-quality synthetic CT scans compared to the conventional Unet.  

This study presents a deep learning based technique for slice super resolution in RADTSE T2 mapping. The proposed method may be used to accelerate full liver imaging, while maintaining sufficient through-plane resolution for detection of small pathologies.

The purpose of this work was to automatically classify BPE using T1-weighted subtraction volumes and diffusion-weighted imaging volumes in breast MRI. The dataset consisted of 621 routine breast MRI examination acquired at University Hospital Erlangen. 2D MIP and 3D T1-subtraction volumes were used for the automatic detection of BPE classes. Multi-b-value DWI (up to1500s/mm2) DWI images were used for automatic prediction. ResNet and DenseNet models were used for 2D and 3D data respectively. The study demonstrated an AUROC of 0.8107 on the test set using the T1-subtraction volumes. With DWI volumes, a slightly decreased AuROC of 0.78 was achieved.

Compressed sensing allowed speeding up MRI image acquisition by several folds but with increased reconstruction time. Deep-learning was introduced for fast reconstruction with comparable or improved reconstruction quality. However, its applications on non-cartesian grids are scarce, mostly based on simulated resampling from original Cartesian grids. Here, we evaluated the performance of two deep-learning networks for reconstructing images acquired with k-space spiral trajectories in real MRI scanning. We demonstrated that deep learning networks can successfully reconstruct high-resolution images from under-sampled spiral trajectories with half of data in k-space and their performance is robust to spiral trajectories different from training.