A model-based MR parameter mapping network robust to substantial variations in acquisition settings.

Deep learning methods show great potential for the efficient and precise estimation of quantitative parameter maps from multiple magnetic resonance (MR) images. Current deep learning-based MR parameter mapping (MPM) methods are mostly trained and tested using data with specific acquisition settings. However, scan protocols usually vary with centers, scanners, and studies in practice. Thus, deep learning methods applicable to MPM with varying acquisition settings are highly required but still rarely investigated. In this work, we develop a model-based deep network termed MMPM-Net for robust MPM with varying acquisition settings. A deep learning-based denoiser is introduced to construct the regularization term in the nonlinear inversion problem of MPM. The alternating direction method of multipliers is used to solve the optimization problem and then unrolled to construct MMPM-Net. The variation in acquisition parameters can be addressed by the data fidelity component in MMPM-Net. Extensive experiments are performed on R2 mapping and R1 mapping datasets with substantial variations in acquisition settings, and the results demonstrate that the proposed MMPM-Net method outperforms other state-of-the-art MR parameter mapping methods both qualitatively and quantitatively.

A fully automatic parenchyma extraction method for MRI T2* relaxometry of iron loaded liver in transfusion-dependent patients.

PURPOSE: To develop a fully automatic parenchyma extraction method for the T2* relaxometry of iron overload liver. METHODS: A retrospective multicenter collection of liver MR examinations from 177 transfusion-dependent patients was conducted. The proposed method extended a semiautomatic parenchyma extraction algorithm to a fully automatic approach by introducing a modified TransUNet on the R2* (1/T2*) map for liver segmentation. Axial liver slices from 129 patients at 1.5 T were allocated to training (85%) and internal test (15%) sets. Two external test sets separately included 1.5 T data from 20 patients and 3.0 T data from 28 patients. The final T2* measurement was obtained by fitting the average signal of the extracted liver parenchyma. The agreement between T2* measurements using fully and semiautomatic parenchyma extraction methods was assessed using coefficient of variation (CoV) and Bland-Altman plots. RESULTS: Dice of the deep network-based liver segmentation was 0.970 ± 0.019 on the internal dataset, 0.960 ± 0.035 on the external 1.5 T dataset, and 0.958 ± 0.014 on the external 3.0 T dataset. The mean difference bias between T2* measurements of the fully and semiautomatic methods were separately 0.12 (95% CI: -0.37, 0.61) ms, 0.04 (95% CI: -1.0, 1.1) ms, and 0.01 (95% CI: -0.25, 0.23) ms on the three test datasets. The CoVs between the two methods were 4.2%, 4.8% and 2.0% on the internal test set and two external test sets. CONCLUSIONS: The developed fully automatic parenchyma extraction approach provides an efficient and operator-independent T2* measurement for assessing hepatic iron content in clinical practice.

MRI Deep Learning-Based Automatic Segmentation of Interventricular Septum for Black-Blood Myocardial T2* Measurement in Thalassemia.

BACKGROUND: The T2* value of interventricular septum is routinely reported for grading myocardial iron load in thalassemia major, and automatic segmentation of septum could shorten analysis time and reduce interobserver variability. PURPOSE: To develop a deep learning-based method for automatic septum segmentation from black-blood MR images for the myocardial T2* measurement of thalassemia patients. STUDY TYPE: Retrospective. POPULATION/SUBJECTS: One hundred forty-six transfusion-dependent thalassemia patients with cardiac MR examinations from two centers. Data from Center 1 (1.5 T) were assigned to the training (100 examinations) and internal testing (20 examinations) sets; data from Center 2 were assigned to the external testing set (26 examinations; 10 at 1.5 T and 16 at 3.0 T). FIELD STRENGTH/SEQUENCE: 1.5 T and 3.0 T, multiecho gradient-echo sequence. ASSESSMENT: A modified attention U-Net for septum segmentation was constructed and trained, and its performance evaluated on unseen internal and external datasets. T2* was measured by fitting the average septum signal, separately segmented by automatic and manual methods. STATISTICAL TESTS: Agreement between manual and automatic septum segmentations was assessed with the Dice coefficient, and T2* agreement was assessed using the Bland-Altman plot and the coefficient of variation (CoV). RESULTS: The median Dice coefficient of deep network-based septum segmentation was 0.90 [0.05] on the internal dataset, 0.82 [0.10] on the external 1.5 T dataset, and 0.86 [0.14] on the external 3.0 T dataset. T2* measurements using automatic segmentation corresponded with those from manual segmentation, with a mean difference of 0.02 (95% LoA: -0.74 to 0.79) msec, 0.43 (95% LoA: -2.1 to 3.0) msec, and 0.36 (95% LoA: -0.72 to 1.4) msec on the three datasets. The CoVs between the two methods were 3.1%, 7.0%, and 6.1% on the internal and two external datasets, respectively. DATA CONCLUSIONS: The proposed septum segmentation yielded myocardial T2* measurements which were highly consistent with those obtained by manual segmentation. This automatic approach may facilitate data processing and avoid operator-dependent variability in practice. EVIDENCE LEVEL: 4 TECHNICAL EFFICACY: Stage 1.

Cascade of Denoising and Mapping Neural Networks for MRI R2* Relaxometry of Iron-Loaded Liver.

MRI of effective transverse relaxation rate (R2*) measurement is a reliable method for liver iron concentration quantification. However, R2* mapping can be degraded by noise, especially in the case of iron overload. This study aimed to develop a deep learning method for MRI R2* relaxometry of an iron-loaded liver using a two-stage cascaded neural network. The proposed method, named CadamNet, combines two convolutional neural networks separately designed for image denoising and parameter mapping into a cascade framework, and the physics-based R2* decay model was incorporated in training the mapping network to enforce data consistency further. CadamNet was trained using simulated liver data with Rician noise, which was constructed from clinical liver data. The performance of CadamNet was quantitatively evaluated on simulated data with varying noise levels as well as clinical liver data and compared with the single-stage parameter mapping network (MappingNet) and two conventional model-based R2* mapping methods. CadamNet consistently achieved high-quality R2* maps and outperformed MappingNet at varying noise levels. Compared with conventional R2* mapping methods, CadamNet yielded R2* maps with lower errors, higher quality, and substantially increased efficiency. In conclusion, the proposed CadamNet enables accurate and efficient iron-loaded liver R2* mapping, especially in the presence of severe noise.