A deep learning approach for magnetic resonance fingerprinting: Scaling capabilities and good training practices investigated by simulations

Fine-Tuning Deep Learning Model for Quantitative Knee Joint Mapping With MR Fingerprinting and Its Comparison to Dictionary Matching Method: Fine-Tuning Deep Learning Model for Quantitative MRF

Magnetic resonance fingerprinting (MRF), as an emerging versatile and noninvasive imaging technique, provides simultaneous quantification of multiple quantitative MRI parameters, which have been used to detect changes in cartilage composition and structure in osteoarthritis. Deep learning (DL)-based methods for quantification mapping in MRF overcome the memory constraints and offer faster processing compared to the conventional dictionary matching (DM) method. However, limited attention has been given to the fine-tuning of neural networks (NNs) in DL and fair comparison with DM. In this study, we investigate the impact of training parameter choices on NN performance and compare the fine-tuned NN with DM for multiparametric mapping in MRF. Our approach includes optimizing NN hyperparameters, analyzing the singular value decomposition (SVD) components of MRF data, and optimization of the DM method. We conducted experiments on synthetic data, the NIST/ISMRM MRI system phantom with ground truth, and in vivo knee data from 14 healthy volunteers. The results demonstrate the critical importance of selecting appropriate training parameters, as these significantly affect NN performance. The findings also show that NNs improve the accuracy and robustness of T1, T2, and T1ρ mappings compared to DM in synthetic datasets. For in vivo knee data, the NN achieved comparable results for T1, with slightly lower T2 and slightly higher T1ρ measurements compared to DM. In conclusion, the fine-tuned NN can be used to increase accuracy and robustness for multiparametric quantitative mapping from MRF of the knee joint.

Improving Accuracy and Reproducibility of Cartilage T2 Mapping in the OAI Dataset Through Extended Phase Graph Modeling

BACKGROUND

The Osteoarthritis Initiative (OAI) collected extensive imaging data, including Multi-Echo Spin-Echo (MESE) sequences for measuring knee cartilage T2 relaxation times. Mono-exponential models are used in the OAI for T2 fitting, which neglects stimulated echoes and B1 inhomogeneities. Extended Phase Graph (EPG) modeling addresses these limitations but has not been applied to the OAI dataset.

PURPOSE

To assess how different fitting methods, including EPG-based and exponential-based approaches, affect the accuracy and reproducibility of cartilage T2 in the OAI dataset.

STUDY TYPE

Retrospective.

POPULATION

From OAI dataset, 50 subjects, stratified by osteoarthritis (OA) severity using Kellgren-Lawrence grades (KLG), and 50 subjects without OA diagnosis during OAI duration were selected (each group: 25 females, mean ages ~61 years).

FIELD STRENGTH/SEQUENCE

3-T, two-dimensional (2D) MESE sequence.

ASSESSMENT

Femoral and tibial cartilages were segmented from DESS images, subdivided into seven sub-regions, and co-registered to MESE. T2 maps were obtained using three EPG-based methods (nonlinear least squares, dictionary matching, and deep learning) and three mono-exponential approaches (linear least squares, nonlinear least squares, and noise-corrected exponential). Average T2 values within sub-regions were obtained. Pair-wise agreement among fitting methods was evaluated using the stratified subjects, while reproducibility using healthy subjects. Each method's T2 accuracy and repeatability varying signal-to-noise ratio (SNR) were assessed with simulations.

STATISTICAL TESTS

Bland-Altman analysis, Lin's concordance coefficient, and coefficient of variation assessed agreement, repeatability, and reproducibility. Statistical significance was set at P-value <0.05.

RESULTS

EPG-based methods demonstrated superior T2 accuracy (mean absolute error below 0.5 msec at SNR > 100) compared to mono-exponential methods (error > 7 msec). EPG-based approaches had better reproducibility, with limits of agreement 1.5-5 msec narrower than exponential-based methods. T2 values from EPG methods were systematically 10-17 msec lower than those from mono-exponential fitting.

DATA CONCLUSION

EPG modeling improved agreement and reproducibility of cartilage T2 mapping in subjects from the OAI dataset.

EVIDENCE LEVEL

3 TECHNICAL EFFICACY: Stage 1.

Microstructure-Informed Myelin Mapping (MIMM) from routine multi-echo gradient echo data using multiscale physics modeling of iron and myelin effects and QSM

PURPOSE

Myelin quantification is used in the study of demyelination in neurodegenerative diseases. A novel noninvasive MRI method, Microstructure-Informed Myelin Mapping (MIMM), is proposed to quantify the myelin volume fraction (MVF) from a routine multi-gradient echo sequence (mGRE) using a multiscale biophysical signal model of the effects of microstructural myelin and iron.

THEORY AND METHODS

In MIMM, the effects of myelin are modeled based on the Hollow Cylinder Fiber Model accounting for anisotropy, while iron is considered as an isotropic paramagnetic point source. This model is used to create a dictionary of mGRE magnitude signal evolution and total voxel susceptibility using finite elements of size 0.2 μm. Next, voxel-by-voxel stochastic matching pursuit between acquired mGRE data (magnitude+QSM) and the pre-computed dictionary generates quantitative MVF and iron susceptibility maps. Dictionary matching was evaluated under three conditions: (1) without fiber orientation (basic), (2) with fiber orientation obtained using DTI, and (3) with fiber orientation obtained using an atlas (atlas). MIMM was compared with the three-pool complex fitting (3PCF) using T2-relaxometry myelin water fraction (MWF) map as reference.

RESULTS

The DTI MIMM and atlas MIMM approaches were equally effective in reducing the overestimation of MVF in certain white matter tracts observed in the basic MIMM approach, and they both showed good agreement with T2-relaxometry MWF. MIMM MVF reduced myelin overestimation of globus pallidus observed in 3PCF MWF.

CONCLUSION

MIMM processing of mGRE data can provide MVF maps from routine clinical scans without requiring special sequences.

Synthesis of MR fingerprinting information from magnitude-only MR imaging data using a parallelized, multi network U-Net convolutional neural network

Background

MR fingerprinting (MRF) is a novel method for quantitative assessment of in vivo MR relaxometry that has shown high precision and accuracy. However, the method requires data acquisition using customized, complex acquisition strategies and dedicated post processing methods thereby limiting its widespread application.

Objective

To develop a deep learning (DL) network for synthesizing MRF signals from conventional magnitude-only MR imaging data and to compare the results to the actual MRF signal acquired.

Methods

A U-Net DL network was developed to synthesize MRF signals from magnitude-only 3D T 1-weighted brain MRI data acquired from 37 volunteers aged between 21 and 62 years of age. Network performance was evaluated by comparison of the relaxometry data (T 1, T 2) generated from dictionary matching of the deep learning synthesized and actual MRF data from 47 segmented anatomic regions. Clustered bootstrapping involving 10,000 bootstraps followed by calculation of the concordance correlation coefficient were performed for both T 1 and T 2 MRF data pairs. 95% confidence limits and the mean difference between true and DL relaxometry values were also calculated.

Results

The concordance correlation coefficient (and 95% confidence limits) for T 1 and T 2 MRF data pairs over the 47 anatomic segments were 0.8793 (0.8136-0.9383) and 0.9078 (0.8981-0.9145) respectively. The mean difference (and 95% confidence limits) were 48.23 (23.0-77.3) s and 2.02 (-1.4 to 4.8) s.

Conclusion

It is possible to synthesize MRF signals from MRI data using a DL network, thereby creating the potential for performing quantitative relaxometry assessment without the need for a dedicated MRF pulse sequence.

A deep learning approach for fast muscle water T2 mapping with subject specific fat T2 calibration from multi-spin-echo acquisitions

This work presents a deep learning approach for rapid and accurate muscle water T2 with subject-specific fat T2 calibration using multi-spin-echo acquisitions. This method addresses the computational limitations of conventional bi-component Extended Phase Graph fitting methods (nonlinear-least-squares and dictionary-based) by leveraging fully connected neural networks for fast processing with minimal computational resources. We validated the approach through in vivo experiments using two different MRI vendors. The results showed strong agreement of our deep learning approach with reference methods, summarized by Lin's concordance correlation coefficients ranging from 0.89 to 0.97. Further, the deep learning method achieved a significant computational time improvement, processing data 116 and 33 times faster than the nonlinear least squares and dictionary methods, respectively. In conclusion, the proposed approach demonstrated significant time and resource efficiency improvements over conventional methods while maintaining similar accuracy. This methodology makes the processing of water T2 data faster and easier for the user and will facilitate the utilization of the use of a quantitative water T2 map of muscle in clinical and research studies.

Fast deep learning reconstruction techniques for preclinical magnetic resonance fingerprinting

We propose a deep learning (DL) model and a hyperparameter optimization strategy to reconstruct T1 and T2 maps acquired with the magnetic resonance fingerprinting (MRF) methodology. We applied two different MRF sequence routines to acquire images of ex vivo rat brain phantoms using a 7-T preclinical scanner. Subsequently, the DL model was trained using experimental data, completely excluding the use of any theoretical MRI signal simulator. The best combination of the DL parameters was implemented by an automatic hyperparameter optimization strategy, whose key aspect is to include all the parameters to the fit, allowing the simultaneous optimization of the neural network architecture, the structure of the DL model, and the supervised learning algorithm. By comparing the reconstruction performances of the DL technique with those achieved from the traditional dictionary-based method on an independent dataset, the DL approach was shown to reduce the mean percentage relative error by a factor of 3 for T1 and by a factor of 2 for T2 , and to improve the computational time by at least a factor of 37. Furthermore, the proposed DL method enables maintaining comparable reconstruction performance, even with a lower number of MRF images and a reduced k-space sampling percentage, with respect to the dictionary-based method. Our results suggest that the proposed DL methodology may offer an improvement in reconstruction accuracy, as well as speeding up MRF for preclinical, and in prospective clinical, investigations.

An interpretable feature-learned model for overall survival classification of High-Grade Gliomas

PURPOSE

An accurate and well-defined survival prediction of High Grade Gliomas (HGGs) is indispensable because of its high incidence and aggressiveness. Therefore, this paper presents a unified framework for fully automatic overall survival classification and its interpretation.

METHODS AND MATERIALS

Initially, a glioma detection model is utilized to detect the tumorous images. A pre-processing module is designed for extracting 2D slices and creating a survival data array for the classification network. Then, the classification pipeline is integrated with two separate pathways: a modality-specific and a modality-concatenated pathway. The modality-specific pathway runs three separate CNNs for extracting rich predictive features from three sub-regions of HGGs (peritumoral edema, enhancing tumor and necrosis) by using three neuro-imaging modalities. In these pathways, the image vectors of the different modalities are also concatenated to the final fusion layer to overcome the loss of lower-level tumor features. Furthermore, to exploit the intra-modality correlations, a modality-concatenated pathway is also added to the classification pipeline. The experiments are conducted on BraTS 2018 and BraTS 2019 benchmarks, demonstrating that the proposed approach performs competitively in classifying HGG patients into three survival groups, namely, short, mid, and long survivors.

RESULTS

The proposed approach achieves an overall classification accuracy, sensitivity, and specificity of about 0.998, 0.997, and 0.999, respectively, for the BraTS 2018 dataset, and for BraTS 2019, these values correspond to 1.000, 0.999, and 0.999.

CONCLUSIONS

The results indicate that the proposed model achieves the highest values of the evaluation metrics for the overall survival classification of HGG.

Microstructural parameters from DW-MRI for tumour characterization and local recurrence prediction in particle therapy of skull-base chordoma

BACKGROUND

Quantitative imaging such as Diffusion-Weighted MRI (DW-MRI) can be exploited to non-invasively derive patient-specific tumor microstructure information for tumor characterization and local recurrence risk prediction in radiotherapy.

PURPOSE

To characterize tumor microstructure according to proliferative capacity and predict local recurrence through microstructural markers derived from pre-treatment conventional DW-MRI, in skull-base chordoma (SBC) patients treated with proton (PT) and carbon ion (CIRT) radiotherapy.

METHODS

Forty-eight patients affected by SBC, who underwent conventional DW-MRI before treatment and were enrolled for CIRT (n = 25) or PT (n = 23), were retrospectively selected. Clinically verified local recurrence information (LR) and histological information (Ki-67, proliferation index) were collected. Apparent diffusion coefficient (ADC) maps were calculated from pre-treatment DW-MRI and, from these, a set of microstructural parameters (cellular radius R, volume fraction vf, diffusion D) were derived by applying a fine-tuning procedure to a framework employing Monte Carlo simulations on synthetic cell substrates. In addition, apparent cellularity (ρ_app ) was estimated from vf and R for an easier clinical interpretation. Histogram-based metrics (mean, median, variance, entropy) from estimated parameters were considered to investigate differences (Mann-Whitney U-test, α = 0.05) in estimated tumor microstructure in SBCs characterized by low or high cell proliferation (Ki-67). Recurrence-free survival analyses were also performed to assess the ability of the microstructural parameters to stratify patients according to the risk of local recurrence (Kaplan-Meier curves, log-rank test α = 0.05).

RESULTS

Refined microstructural markers revealed optimal capabilities in discriminating patients according to cell proliferation, achieving best results with mean values (p-values were 0.0383, 0.0284, 0.0284, 0.0468, and 0.0088 for ADC, R, vf, D, and ρ_app, respectively). Recurrence-free survival analyses showed significant differences between populations at high and low risk of local recurrence as stratified by entropy values of estimated microstructural parameters (p = 0.0110).

CONCLUSION

Patient-specific microstructural information was non-invasively derived providing potentially useful tools for SBC treatment personalization and optimization in particle therapy.

Optimisation of data acquisition towards continuous cardiac Magnetic Resonance Fingerprinting applications

PURPOSE

Assess and optimise acquisition parameters for continuous cardiac Magnetic Resonance Fingerprinting (MRF).

METHODS

Different acquisition schemes (flip angle amplitude, lobe size, T2-preparation pulses) for cardiac MRF were assessed in simulations and phantom and demonstrated in one healthy volunteer. Three different experimental designs were evaluated using central composite and fractional factorial designs. Relative errors for T1 and T2 were calculated for a wide range of realistic T1 and T2 value combinations. The effect of different designs on the accuracy of T1 and T2 was assessed using response surface modelling and Cohen's f calculations.

RESULTS

Larger flip angle amplitudes lead to an improvement of T2 accuracy and precision for simulations and phantom experiments. Similar effects could also be shown qualitatively in in-vivo scans. Accuracy and precision of T1 were robust to different design parameters with improved values for faster flip angle variation. Cohen's f showed that T2-preparation pulses influence the accuracy of T2. The number of pulses used is the most important parameter. Without T2-preparation pulses, RMSE were 3.0 ± 8.09 % for T1 and 16.24 ± 14.47 % for T2. Using those pulses reduced the RMSE to 2.3 ± 8.4 % for T1 and 14.11 ± 13.46 % for T2. Nonetheless, even if the improvement is significant, RMSE are still too high for reliable quantification.

CONCLUSION

In contrast to previous study using triggered MRF sequences using < 30° flip angles, large flip angle amplitudes led to better results for continuous cardiac MRF sequences. T2-preparation pulse can improve the accuracy of T2 estimation but lead to longer scan times.

Artificial intelligence in multi-parametric magnetic resonance imaging: A review

Multi-parametric magnetic resonance imaging (mpMRI) is an indispensable tool in the clinical workflow for the diagnosis and treatment planning of various diseases. Machine learning-based artificial intelligence (AI) methods, especially those adopting the deep learning technique, have been extensively employed to perform mpMRI image classification, segmentation, registration, detection, reconstruction, and super-resolution. The current availability of increasing computational power and fast-improving AI algorithms have empowered numerous computer-based systems for applying mpMRI to disease diagnosis, imaging-guided radiotherapy, patient risk and overall survival time prediction, and the development of advanced quantitative imaging technology for magnetic resonance fingerprinting. However, the wide application of these developed systems in the clinic is still limited by a number of factors, including robustness, reliability, and interpretability. This survey aims to provide an overview for new researchers in the field as well as radiologists with the hope that they can understand the general concepts, main application scenarios, and remaining challenges of AI in mpMRI. This article is protected by copyright. All rights reserved.