Repeatability of IVIM biomarkers from diffusion-weighted MRI in head and neck: Bayesian probability versus neural network

Velocity-compensated intravoxel incoherent motion of the human calf muscle

PURPOSE

To determine whether intravoxel incoherent motion (IVIM) describes the blood perfusion in muscles better, assuming pseudo diffusion (Bihan Model 1) or ballistic motion (Bihan Model 2).

METHODS

IVIM parameters were measured in 18 healthy subjects with three different diffusion gradient time profiles (bipolar with two diffusion times and one with velocity compensation) and 17 b-values (0-600 s/mm2) at rest and after muscle activation. The diffusion coefficient, perfusion fraction, and pseudo-diffusion coefficient were estimated with a segmented fit in the gastrocnemius medialis (GM) and tibialis anterior (TA) muscles.

RESULTS

Velocity-compensated gradients resulted in a decreased perfusion fraction (6.9% ± 1.4% vs. 4.4% ± 1.3% in the GM after activation) and pseudo-diffusion coefficient (0.069 ± 0.046 mm2/s vs. 0.014 ± 0.006 in the GM after activation) compared to the bipolar gradients with the longer diffusion encoding time. Increased diffusion coefficients, perfusion fractions, and pseudo-diffusion coefficients were observed in the GM after activation for all gradient profiles. However, the increase was significantly smaller for the velocity-compensated gradients. A diffusion time dependence was found for the pseudo-diffusion coefficient in the activated muscle.

CONCLUSION

Velocity-compensated diffusion gradients significantly suppress the IVIM effect in the calf muscle, indicating that the ballistic limit is mostly reached, which is supported by the time dependence of the pseudo-diffusion coefficient.

Calculation of intravoxel incoherent motion parameter maps using a kernelized total difference-based method

Quantitative analysis of diffusion-weighted magnetic resonance imaging (DW-MRI) has been explored for many clinical applications since its development. In particular, the intravoxel incoherent motion (IVIM) model for DW-MRI has been commonly utilized in various organs. However, because of the presence of excessive noise, the IVIM parameter maps obtained from pixel-wise fitting are often unreliable. In this study, we propose a kernelized total difference-based curve-fitting method to estimate the IVIM parameters. Simulated DW-MRI data at five signal-to-noise ratios (i.e., 10, 20, 30, 50, and 100) and real abdominal DW-MRI data acquired on a 1.5-T MRI scanner with nine b-values (i.e., 0, 10, 25, 50, 100, 200, 300, 400, and 500 s/mm2) and six diffusion-encoding gradient directions were used to evaluate the performance of the proposed method. The results were compared with those obtained by three existing methods: trust-region reflective (TRR) algorithm, Bayesian probability (BP), and deep neural network (DNN). Our simulation results showed that the proposed method outperformed the other three comparing methods in terms of root-mean-square error. Moreover, the proposed method could preserve small details in the estimated IVIM parameter maps. The experimental results showed that, compared with the TRR method, the proposed method as well as the BP (and DNN) method could reduce the overestimation of the pseudodiffusion coefficient and improve the quality of IVIM parameter maps. For all studied abdominal organs except the pancreas, both the proposed method and the BP method could provide IVIM parameter estimates close to the reference values; the former had higher precision. The kernelized total difference-based curve-fitting method has the potential to improve the reliability of IVIM parametric imaging.

Repeatability and reproducibility comparisons of liver IVIM imaging with free-breathing or respiratory-triggered sequences

For liver intravoxel incoherent motion (IVIM) data acquisition, respiratory-triggering (RT) MRI is commonly used, and there are strong motivations to shorten the scan duration. For the same scan duration, more b values or higher numbers of excitations can be allowed for free-breathing (FB) imaging than for RT. We studied whether FB can be used to replace RT when careful IVIM image acquisition and image processing are conducted. MRI data of 22 healthy participants were acquired using a 3.0 T scanner. Diffusion imaging was based on a single-shot spin-echo-type echo-planar sequence and 16 b values of 0, 2, 4, 7, 10, 15, 20, 30, 46, 60, 72, 100, 150, 200, 400, and 600 s/mm2 . Each subject attended two scan sessions with an interval of 10-20 days. For each scan session, a subject was scanned twice, first with RT and then with FB. The mean image acquisition time was 5.4 min for FB and 10.8 min for RT. IVIM parameters were calculated with bi-exponential model segmented fitting with a threshold b value of 60 s/mm2 , and fitting started from b = 2 s/mm2 . There was no statistically significant difference between IVIM parameters measured with FB imaging or RT imaging. Perfusion fraction ICC (intraclass correlation coefficient) for FB imaging and RT imaging in the same scan session was 0.824. For perfusion fraction, wSD (within-subject standard deviation), BA (Bland-Altman) difference, BA 95% limit, and ICC were 0.022, 0.0001, -0.0635~0.0637, and 0.687 for FB and 0.031, 0.0122, -0.0723~0.0967, and 0.611 for RT. For Dslow (×10-3  s/mm2 ), wSD, BA difference, BA 95% limit, and ICC were 0.057, 0.0268, -0.1258~0.1793, and 0.471 for FB and 0.073, -0.0078, -0.2170-0.2014, and <0.4 for RT. The Dfast coefficient of variation was 0.20 for FB imaging and 0.28 for RT imaging. All reproducibility indicators slightly favored FB imaging.

Biomarkers Predictive of Distant Disease-free Survival Derived from Diffusion-weighted Imaging of Breast Cancer

PURPOSE

To investigate whether intravoxel incoherent motion (IVIM) and/or non-Gaussian diffusion parameters are associated with distant disease-free survival (DDFS) in patients with invasive breast cancer.

METHODS

From May 2013 to March 2015, 101 patients (mean age 60.0, range 28-88) with invasive breast cancer were evaluated prospectively. IVIM parameters (flowing blood volume fraction [fIVIM] and pseudodiffusion coefficient [D*]) and non-Gaussian diffusion parameters (theoretical apparent diffusion coefficient [ADC] at a b value of 0 s/mm2 [ADC0] and kurtosis [K]) were estimated using a diffusion-weighted imaging series of 16 b values up to 2500 s/mm2. Shifted ADC values (sADC200-1500) and standard ADC values (ADC0-800) were also calculated. The Kaplan-Meier method was used to generate survival analyses for DDFS, which were compared using the log-rank test. Univariable Cox proportional hazards models were used to assess any associations between each parameter and distant metastasis-free survival.

RESULTS

The median observation period was 80 months (range, 35-92 months). Among the 101 patients, 12 (11.9%) developed distant metastasis, with a median time to metastasis of 79 months (range, 10-92 months). Kaplan-Meier analysis showed that DDFS was significantly shorter in patients with K > 0.98 than in those with K ≤ 0.98 (P = 0.04). Cox regression analysis showed a marginal statistical association between K and distant metastasis-free survival (P = 0.05).

CONCLUSION

Non-Gaussian diffusion may be associated with prognosis in invasive breast cancer. A higher K may be a marker to help identify patients at an elevated risk of distant metastasis, which could guide subsequent treatment.

In vivo Mapping of Cellular Resolution Neuropathology in Brain Ischemia by Diffusion MRI

Non-invasive mapping of cellular pathology can provide critical diagnostic and prognostic information. Recent developments in diffusion MRI have produced new tools for examining tissue microstructure at a level well below the imaging resolution. Here, we report the use of diffusion time ( t )-dependent diffusion kurtosis imaging ( t DKI) to simultaneously assess the morphology and transmembrane permeability of cells and their processes in the context of pathological changes in hypoxic-ischemic brain (HI) injury. Through Monte Carlo simulations and cell culture organoid imaging, we demonstrate feasibility in measuring effective size and permeability changes based on the peak and tail of t DKI curves. In a mouse model of HI, in vivo imaging at 11.7T detects a marked shift of the t DKI peak to longer t in brain edema, suggesting swelling and beading associated with the astrocytic processes and neuronal neurites. Furthermore, we observed a faster decrease of the t DKI tail in injured brain regions, reflecting increased membrane permeability that was associated with upregulated water exchange upon astrocyte activation at acute stage as well as necrosis with disrupted membrane integrity at subacute stage. Such information, unavailable with conventional diffusion MRI at a single t, can predict salvageable tissues. For a proof-of-concept, t DKI at 3T on an ischemic stroke patient suggested increased membrane permeability in the stroke region. This work therefore demonstrates the potential of t DKI for in vivo detection of the pathological changes in microstructural morphology and transmembrane permeability after ischemic injury using a clinically translatable protocol.

Effects of fitting methods, high b-values and image quality on diffusion and perfusion quantification and reproducibility in the calf

PURPOSES

The study aimed to optimize diffusion-weighted imaging (DWI) image acquisition and analysis protocols in calf muscles by investigating the effects of different model-fitting methods, image quality, and use of high b-value and constraints on parameters of interest (POIs). The optimized modeling methods were used to select the optimal combinations of b-values, which will allow shorter acquisition time while achieving the same reliability as that obtained using 16 b-values.

METHODS

Test-retest baseline and high-quality DWI images of ten healthy volunteers were acquired on a 3T MR scanner, using 16 b-values, including a high b-value of 1200 s/mm², and structural T1-weighted images for calf muscle delineation. Three and six different fitting methods were used to derive ADC from monoexponential (ME) model and D_d, f_p, and D_p from intravoxel incoherent motion (IVIM) model, with or without the high b-value. The optimized ME and IVIM models were then used to determine the optimal combinations of b-values, obtainable with the least number of b-values, using the selection criteria of coefficient of variance (CV) ≤10% for all POIs.

RESULTS

The find minimum multivariate algorithm was more flexible and yielded smaller fitting errors. The 2-steps fitting method, with fixed D_d, performed the best for IVIM model. The inclusion of high b-value reduced outliers, while constraints improved 2-steps fitting only.

CONCLUSIONS

The optimal numbers of b-values for ME and IVIM models were nine and six b-values respectively. Test-retest reliability analyses showed that only ADC and D_d were reliable for calf diffusion evaluation, with CVs of 7.22% and 4.09%.

An unsupervised convolutional neural network method for estimation of intravoxel incoherent motion parameters

Objective. Intravoxel incoherent motion (IVIM) imaging obtained by fitting a biexponential model to multiple b-value diffusion-weighted magnetic resonance imaging (DW-MRI) has been shown to be a promising tool for different clinical applications. Recently, several deep neural network (DNN) methods were proposed to generate IVIM imaging. Approach. In this study, we proposed an unsupervised convolutional neural network (CNN) method for estimation of IVIM parameters. We used both simulated and real abdominal DW-MRI data to evaluate the performance of the proposed CNN-based method, and compared the results with those obtained from a non-linear least-squares fit (TRR, trust-region reflective algorithm) and a feed-forward backward-propagation DNN-based method. Main results. The simulation results showed that both the DNN- and CNN-based methods had lower coefficients of variation than the TRR method, but the CNN-based method provided more accurate parameter estimates. The results obtained from real DW-MRI data showed that the TRR method produced many biased IVIM parameter estimates that hit the upper and lower parameter bounds. In contrast, both the DNN- and CNN-based methods yielded less biased IVIM parameter estimates. Overall, the perfusion fraction and diffusion coefficient obtained from the DNN- and CNN-based methods were close to literature values. However, compared with the CNN-based method, both the TRR and DNN-based methods tended to yield increased pseudodiffusion coefficients (55%–180%). Significance. Our preliminary results suggest that it is feasible to estimate IVIM parameters using CNN.

Self-supervised neural network improves tri-exponential intravoxel incoherent motion model fitting compared to least-squares fitting in non-alcoholic fatty liver disease

Recent literature suggests that tri-exponential models may provide additional information and fit liver intravoxel incoherent motion (IVIM) data more accurately than conventional bi-exponential models. However, voxel-wise fitting of IVIM results in noisy and unreliable parameter maps. For bi-exponential IVIM, neural networks (NN) were able to produce superior parameter maps than conventional least-squares (LSQ) generated images. Hence, to improve parameter map quality of tri-exponential IVIM, we developed an unsupervised physics-informed deep neural network (IVIM3-NET). We assessed its performance in simulations and in patients with non-alcoholic fatty liver disease (NAFLD) and compared outcomes with bi-exponential LSQ and NN fits and tri-exponential LSQ fits. Scanning was performed using a 3.0T free-breathing multi-slice diffusion-weighted single-shot echo-planar imaging sequence with 18 b-values. Images were analysed for visual quality, comparing the bi- and tri-exponential IVIM models for LSQ fits and NN fits using parameter-map signal-to-noise ratios (SNR) and adjusted R2. IVIM parameters were compared to histological fibrosis, disease activity and steatosis grades. Parameter map quality improved with bi- and tri-exponential NN approaches, with a significant increase in average parameter-map SNR from 3.38 to 5.59 and 2.45 to 4.01 for bi- and tri-exponential LSQ and NN models respectively. In 33 out of 36 patients, the tri-exponential model exhibited higher adjusted R2 values than the bi-exponential model. Correlating IVIM data to liver histology showed that the bi- and tri-exponential NN outperformed both LSQ models for the majority of IVIM parameters (10 out of 15 significant correlations). Overall, our results support the use of a tri-exponential IVIM model in NAFLD. We show that the IVIM3-NET can be used to improve image quality compared to a tri-exponential LSQ fit and provides promising correlations with histopathology similar to the bi-exponential neural network fit, while generating potentially complementary additional parameters.

Bi-exponential fitting excluding b=0 data improves the scan-rescan stability of liver IVIM parameter measures and particularly so for the perfusion fraction

BACKGROUND

A prerequisite to translating intravoxel incoherent motion (IVIM) imaging into meaningful clinical applications is sufficient scan-rescan reproducibility. This study aims to confirm the hypothesis that IVIM data fitting by not using b=0 images will improve the stability of liver IVIM measurement.

METHODS

Healthy volunteers' liver IVIM images were prospectively acquired using a 1.5-T magnet or a 3.0 T with 16 b-values. Repeatability study subjects were scanned twice during the same session, resulted in 35 paired scans for 35 subjects (11 men, mean age: 41.82 years, range: 32-60 years; 24 women, mean age: 42.67 years, range: 20-71 years). IVIM analysis was performed with full-fitting and segmented-fitting with a threshold b-value of 60 s/mm2, and fitting started from b=0 s/mm2 or from b=2 s/mm2. Reproducibility study subjects were scanned and then rescanned with an interval of 5-18 days, resulted in 20 paired scans for 11 subjects (4 men, mean age: 26.25 years, range: 25-27 years; 7 women, mean age: 25.57 years, range: 24-27 years). IVIM analysis was performed with segmented-fitting with a threshold b-value of 50 s/mm2, and fitting started from b=0 s/mm2 or from b=3 s/mm2.

RESULTS

Fitting without b=0 data generally improved the repeatability and reproducibility for both PF and Dslow, and particularly so for PF. For with b=0 data segmented fitting repeatability, PF had within-subject standard deviation of 0.019, bland-Atman 75% agreement limit of -31.52% to 28.35%, and ICC of 0.647, while these values were 0.009, -20.78% to 16.86%, and 0.837 for without b=0 analysis. Though the repeatability and reproducibility for Dfast generally also improved, they remained suboptimal. Measurement stability was better for repeatability than for reproducibility.

CONCLUSIONS

Scan-rescan repeatability and reproducibility of liver IVIM parameters can be improved by fitting without b=0 data, which is particularly so for PF.

Early Response Prediction of Multiparametric Functional MRI and 18F-FDG-PET in Patients with Head and Neck Squamous Cell Carcinoma Treated with (Chemo)Radiation

Simple Summary

Patients with locally-advanced head and neck squamous cell carcinoma (HNSCC) have variable responses to (chemo)radiotherapy. A reliable early prediction of outcomes allows for enhancing treatment efficacy and follow-up monitoring. Early tumoral changes can be captured by functional imaging (DWI/IVIM/DCE/¹⁸F-FDG-PET-CT) parameters, which allow for the construction of accurate patient-specific prognostic models for locoregional recurrence-free survival, distant metastasis-free survival and overall survival. We also present clinical applicable risk stratification in high/medium/low risks for these patient outcomes. This can enable personalized treatment (adaptation) management early on during treatment, improve counseling and enhance patient-specific post-therapy monitoring.

Abstract

Background: Patients with locally-advanced head and neck squamous cell carcinoma (HNSCC) have variable responses to (chemo)radiotherapy. A reliable prediction of outcomes allows for enhancing treatment efficacy and follow-up monitoring. Methods: Fifty-seven histopathologically-proven HNSCC patients with curative (chemo)radiotherapy were prospectively included. All patients had an MRI (DW,-IVIM, DCE-MRI) and ¹⁸F-FDG-PET/CT before and 10 days after start-treatment (intratreatment). Primary tumor functional imaging parameters were extracted. Univariate and multivariate analysis were performed to construct prognostic models and risk stratification for 2 year locoregional recurrence-free survival (LRFFS), distant metastasis-free survival (DMFS) and overall survival (OS). Model performance was measured by the cross-validated area under the receiver operating characteristic curve (AUC). Results: The best LRFFS model contained the pretreatment imaging parameters ADC_kurtosis, K_ep and SUV_peak, and intratreatment imaging parameters change (Δ) Δ-ADC_skewness, Δ-f, Δ-SUV_peak and Δ-total lesion glycolysis (TLG) (AUC = 0.81). Clinical parameters did not enhance LRFFS prediction. The best DMFS model contained pretreatment ADC_kurtosis and SUV_peak (AUC = 0.88). The best OS model contained gender, HPV-status, N-stage, pretreatment ADC_skewness, D, f, metabolic-active tumor volume (MATV), SUV_mean and SUV_peak (AUC = 0.82). Risk stratification in high/medium/low risk was significantly prognostic for LRFFS (p = 0.002), DMFS (p < 0.001) and OS (p = 0.003). Conclusions: Intratreatment functional imaging parameters capture early tumoral changes that only provide prognostic information regarding LRFFS. The best LRFFS model consisted of pretreatment, intratreatment and Δ functional imaging parameters; the DMFS model consisted of only pretreatment functional imaging parameters, and the OS model consisted ofHPV-status, gender and only pretreatment functional imaging parameters. Accurate clinically applicable risk stratification calculators can enable personalized treatment (adaptation) management, early on during treatment, improve counseling and enhance patient-specific post-therapy monitoring.

Usefulness of diffusion derived vessel density computed from a simplified IVIM imaging protocol: An experimental study with rat biliary duct blockage induced liver fibrosis

OBJECTIVES

Liver vessel density can be evaluated by DDVD (diffusion derived vessel density): DDVD_(b0b1) = Sb0/ROIarea0 - Sb1/ROIarea1, where Sb0 and Sb1 refer to the liver signal when b is 0 or 1 s/mm². Sb1 and ROIarea1 may be replaced by other b-values. With a rat biliary duct ligation (BDL) model, this study assesses the usefulness of liver DDVD computed from a simplified IVIM imaging protocol using b = 25 and b = 50 to replace b = 1 s/mm², alone and in combination with other IVIM parameters.

METHODS

Male Sprague-Dawley rats were used. The rat number was 5, 5, 5, and 3 respectively, for the timepoints of 7, 14, 21, 28 days post-BDL surgery. 12 rats had partial biliary duct recanalization performed after the rats had BDL for 7 days and then again followed-up for a mean of 14 days. Liver diffusion MRIs were acquired at 3.0 T with a b-value distribution of 0, 25, 50, 75, 100, 150, 300, 700, 1000 s/mm². DDVD_mean (control rats n = 6) was the mean of DDVD_(b0b25) and DDVD_(b0b50). IVIM fitting started from b = 0 s/mm² with segmented fitting and a threshold b of 50 s/mm² (n = 5 for control rats). Three 3-D spaces were constructed using a combination of the four diffusion parameters.

RESULTS

The control rats and BDL rats (n = 18) had a liver DDVD_mean of 84.0 ± 26.2 and 44.7 ± 14.4 au/pixel (p < 0.001). All 3-D spaces totally separated healthy livers and all fibrotic livers (n = 30, BDL rats and recanalization rats). The mean relative distance between healthy liver cluster and fibrotic liver cluster was 0.331 for PF, D_slow, and D_fast; 0.381 for PF, D_fast, and DDVD_mean; and 0.384 for PF, D_slow, and DDVD_mean.

CONCLUSION

A combination of PF, D_slow, and D_fast allows total separation of healthy livers and fibrotic livers and the integration of DDVD improved the separation.

Intravoxel Incoherent Motion Magnetic Resonance Imaging in Skeletal Muscle: Review and Future Directions

Throughout the body, muscle structure and function can be interrogated using a variety of noninvasive magnetic resonance imaging (MRI) methods. Recently, intravoxel incoherent motion (IVIM) MRI has gained momentum as a method to evaluate components of blood flow and tissue diffusion simultaneously. Much of the prior research has focused on highly vascularized organs, including the brain, kidney, and liver. Unique aspects of skeletal muscle, including the relatively low perfusion at rest and large dynamic range of perfusion between resting and maximal hyperemic states, may influence the acquisition, postprocessing, and interpretation of IVIM data. Here, we introduce several of those unique features of skeletal muscle; review existing studies of IVIM in skeletal muscle at rest, in response to exercise, and in disease states; and consider possible confounds that should be addressed for muscle-specific evaluations. Most studies used segmented nonlinear least squares fitting with a b-value threshold of 200 sec/mm² to obtain IVIM parameters of perfusion fraction (f), pseudo-diffusion coefficient (D*), and diffusion coefficient (D). In healthy individuals, across all muscles, the average ± standard deviation of D was 1.46 ± 0.30 × 10^-3  mm² /sec, D* was 29.7 ± 38.1 × 10^-3  mm² /sec, and f was 11.1 ± 6.7%. Comparisons of reported IVIM parameters in muscles of the back, thigh, and leg of healthy individuals showed no significant difference between anatomic locations. Throughout the body, exercise elicited a positive change of all IVIM parameters. Future directions including advanced postprocessing models and potential sequence modifications are discussed. LEVEL OF EVIDENCE: 2 TECHNICAL EFFICACY: Stage 2.

Repeatability of IVIM biomarkers from diffusion-weighted MRI in head and neck: Bayesian probability versus neural network

Purpose

The intravoxel incoherent motion (IVIM) model for DWI might provide useful biomarkers for disease management in head and neck cancer. This study compared the repeatability of three IVIM fitting methods to the conventional nonlinear least‐squares regression: Bayesian probability estimation, a recently introduced neural network approach, IVIM‐NET, and a version of the neural network modified to increase consistency, IVIM‐NET_mod.

Methods

Ten healthy volunteers underwent two imaging sessions of the neck, two weeks apart, with two DWI acquisitions per session. Model parameters (ADC, diffusion coefficient Dt, perfusion fraction fp, and pseudo‐diffusion coefficient Dp) from each fit method were determined in the tonsils and in the pterygoid muscles. Within‐subject coefficients of variation (wCV) were calculated to assess repeatability. Training of the neural network was repeated 100 times with random initialization to investigate consistency, quantified by the coefficient of variance.

Results

The Bayesian and neural network approaches outperformed nonlinear regression in terms of wCV. Intersession wCV of Dt in the tonsils was 23.4% for nonlinear regression, 9.7% for Bayesian estimation, 9.4% for IVIM‐NET, and 11.2% for IVIM‐NET_mod. However, results from repeated training of the neural network on the same data set showed differences in parameter estimates: The coefficient of variances over the 100 repetitions for IVIM‐NET were 15% for both Dt and fp, and 94% for Dp; for IVIM‐NET_mod, these values improved to 5%, 9%, and 62%, respectively.

Conclusion

Repeatabilities from the Bayesian and neural network approaches are superior to that of nonlinear regression for estimating IVIM parameters in the head and neck.