Development and stability analysis of carpal kinematic metrics from 4D magnetic resonance imaging.

OBJECTIVE: To develop MRI-derived carpal kinematic metrics and investigating their stability. METHODS: The study used a 4D MRI method to track scaphoid, lunate, and capitate movements in the wrist. A panel of 120 metrics for radial-ulnar deviation and flexion-extension was created using polynomial models of scaphoid and lunate movements relative to the capitate. Intraclass correlation coefficients (ICCs) analyzed intra- and inter-subject stability in 49 subjects, 20 with and 29 without wrist injury history. RESULTS: Comparable degrees of stability were observed across the two different wrist movements. Among the total 120 derived metrics, distinct subsets demonstrated high stability within each type of movement. For asymptomatic subjects, 16 out of 17 metrics with high intra-subject stability also showed high inter-subject stability. The differential analysis of ICC values for each metric between asymptomatic and symptomatic cohorts revealed specific metrics (although relatively unstable) exhibiting greater variability in the symptomatic cohort, thereby highlighting the impact of wrist conditions on the variability of kinematic metrics. CONCLUSION: The findings demonstrate the developing potential of dynamic MRI for assessing and characterizing complex carpal bone dynamics. Stability analyses of the derived kinematic metrics revealed encouraging differences between cohorts with and without wrist injury histories. Although these broad metric stability variations highlight the potential utility of this approach for analyzing carpal instability, further studies are necessary to better characterize these observations.

Split-slice training and hyperparameter tuning of RAKI networks for simultaneous multi-slice reconstruction.

PURPOSE: Simultaneous multi-slice acquisitions are essential for modern neuroimaging research, enabling high temporal resolution functional and high-resolution q-space sampling diffusion acquisitions. Recently, deep learning reconstruction techniques have been introduced for unaliasing these accelerated acquisitions, and robust artificial-neural-networks for k-space interpolation (RAKI) have shown promising capabilities. This study systematically examines the impacts of hyperparameter selections for RAKI networks, and introduces a novel technique for training data generation which is analogous to the split-slice formalism used in slice-GRAPPA. METHODS: RAKI networks were developed with variable hyperparameters and with and without split-slice training data generation. Each network was trained and applied to five different datasets including acquisitions harmonized with Human Connectome Project lifespan protocol. Unaliasing performance was assessed through L1 errors computed between unaliased and calibration frequency-space data. RESULTS: Split-slice training significantly improved network performance in nearly all hyperparameter configurations. Best unaliasing results were achieved with three layer RAKI networks using at least 64 convolutional filters with receptive fields of 7 voxels, 128 single-voxel filters in the penultimate RAKI layer, batch normalization, and no training dropout with the split-slice augmented training dataset. Networks trained without the split-slice technique showed symptoms of network over-fitting. CONCLUSIONS: Split-slice training for simultaneous multi-slice RAKI networks positively impacts network performance. Hyperparameter tuning of such reconstruction networks can lead to further improvements in unaliasing performance.

Optimization of q-space sampling for mean apparent propagator MRI metrics using a genetic algorithm.

Mean Apparent Propagator (MAP) MRI is a recently introduced technique to estimate the diffusion probability density function (PDF) robustly. Using the estimated PDF, MAP MRI then calculates zero-displacement and non-Gaussianity metrics, which might better characterize tissue microstructure compared to diffusion tensor imaging or diffusion kurtosis imaging. However, intensive q-space sampling required for MAP MRI limits its widespread adoption. A reduced q-space sampling scheme that maintains the accuracy of the derived metrics would make it more practical. A heuristic approach for acquiring MAP MRI with fewer q-space samples has been introduced earlier with scan duration of less than 10 minutes. However, the sampling scheme was not optimized systematically to preserve the accuracy of the model metrics. In this work, a genetic algorithm is implemented to determine optimal q-space subsampling schemes for MAP MRI that will keep total scan time under 10 min. Results show that the metrics derived from the optimized schemes more closely match those computed from the full set, especially in dense fiber tracts such as the corpus callosum.

Analysis of errors in diffusion kurtosis imaging caused by slice crosstalk in simultaneous multi-slice imaging.

Simultaneous multi-slice (SMS) imaging techniques accelerate diffusion MRI data acquisition. However, slice separation is imperfect and results in residual signal leakage between the simultaneously excited slices. The resulting consistent bias may adversely affect diffusion model parameter estimation. Although this bias is usually small and might not affect the simplified diffusion tensor model significantly, higher order diffusion models such as kurtosis are likely to be more susceptible to such effects. In this work, two SMS reconstruction techniques and an alternative acquisition approach were tested to quantify the effects of slice crosstalk on diffusion kurtosis parameters. In reconstruction, two popular slice separation algorithms, slice GRAPPA and split-slice GRAPPA, are evaluated to determine the effect of slice leakage on diffusion kurtosis metrics. For the alternative acquisition, the slice pairings were varied across diffusion weighted images such that the signal leakage does not come from the same overlapped slice for all diffusion encodings. Simulation results demonstrated the potential benefits of randomizing the slice pairings. However, various experimental factors confounded the advantages of slice pair randomization. In volunteer experiments, region-of-interest analyses found high metric errors with each of the SMS acquisitions and reconstructions in the brain white matter.

Assessing diffusion kurtosis tensor estimation methods using a digital brain phantom derived from human connectome project data.

PURPOSE: Diffusion kurtosis imaging (DKI) has gained popularity in recent years as an advanced diffusion-weighted MRI technique. This work aims to quantitatively compare the performance and accuracy of four DKI processing algorithms. For this purpose, a digital DKI brain phantom is developed. METHODS: Data from the Human Connectome Project database were used to generate a DKI digital phantom. In a Monte Carlo Rician noise simulation, four DKI processing algorithms were compared based on their mean squared error, squared bias, and variance. RESULTS: Algorithm performance was region-dependent and differed for each diffusion metric and noise level. Crossover between variance and squared bias error occurred between signal-to-noise ratios of 30 and 40. CONCLUSION: Through the framework presented here, DKI algorithms can be quantitatively compared via a ground truth data set. Error maps are critical as algorithm performance varies spatially. Bias-plus-variance decomposition provides a more complete picture than MSE alone. In combination with refinements in acquisition in future studies, the accuracy and efficiency of DKI will continue to improve promoting clinical adoption.