Radiological features of brain hemorrhage through automated segmentation from computed tomography in stroke and traumatic brain injury.

Introduction: Radiological assessment is necessary to diagnose spontaneous intracerebral hemorrhage (ICH) and traumatic brain injury intracranial hemorrhage (TBI-bleed). Artificial intelligence (AI) deep learning tools provide a means for decision support. This study evaluates the hemorrhage segmentations produced from three-dimensional deep learning AI model that was developed using non-contrast computed tomography (CT) imaging data external to the current study. Methods: Non-contrast CT imaging data from 1263 patients were accessed across seven data sources (referred to as sites) in Norway and Sweden. Patients were included based on ICH, TBI-bleed, or mild TBI diagnosis. Initial non-contrast CT images were available for all participants. Hemorrhage location frequency maps were generated. The number of estimated haematoma clusters was correlated with the total haematoma volume. Ground truth expert annotations were available for one ICH site; hence, a comparison was made with the estimated haematoma volumes. Segmentation volume estimates were used in a receiver operator characteristics (ROC) analysis for all samples (i.e., bleed detected) and then specifically for one site with few TBI-bleed cases. Results: The hemorrhage frequency maps showed spatial patterns of estimated lesions consistent with ICH or TBI-bleed presentations. There was a positive correlation between the estimated number of clusters and total haematoma volume for each site (correlation range: 0.45-0.74; each p-value < 0.01) and evidence of ICH between-site differences. Relative to hand-drawn annotations for one ICH site, the VIOLA-AI segmentation mask achieved a median Dice Similarity Coefficient of 0.82 (interquartile range: 0.78 and 0.83), resulting in a small overestimate in the haematoma volume by a median of 0.47 mL (interquartile range: 0.04 and 1.75 mL). The bleed detection ROC analysis for the whole sample gave a high area-under-the-curve (AUC) of 0.92 (with sensitivity and specificity of 83.28% and 95.41%); however, when considering only the mild head injury site, the TBI-bleed detection gave an AUC of 0.70. Discussion: An open-source segmentation tool was used to visualize hemorrhage locations across multiple data sources and revealed quantitative hemorrhage site differences. The automated total hemorrhage volume estimate correlated with a per-participant hemorrhage cluster count. ROC results were moderate-to-high. The VIOLA-AI tool had promising results and might be useful for various types of intracranial hemorrhage.

A densely interconnected network for deep learning accelerated MRI.

OBJECTIVE: To improve accelerated MRI reconstruction through a densely connected cascading deep learning reconstruction framework. MATERIALS AND METHODS: A cascading deep learning reconstruction framework (reference model) was modified by applying three architectural modifications: input-level dense connections between cascade inputs and outputs, an improved deep learning sub-network, and long-range skip-connections between subsequent deep learning networks. An ablation study was performed, where five model configurations were trained on the NYU fastMRI neuro dataset with an end-to-end scheme conjunct on four- and eightfold acceleration. The trained models were evaluated by comparing their respective structural similarity index measure (SSIM), normalized mean square error (NMSE), and peak signal to noise ratio (PSNR). RESULTS: The proposed densely interconnected residual cascading network (DIRCN), utilizing all three suggested modifications achieved a SSIM improvement of 8% and 11%, a NMSE improvement of 14% and 23%, and a PSNR improvement of 2% and 3% for four- and eightfold acceleration, respectively. In an ablation study, the individual architectural modifications all contributed to this improvement for both acceleration factors, by improving the SSIM, NMSE, and PSNR with approximately 2-4%, 4-9%, and 0.5-1%, respectively. CONCLUSION: The proposed architectural modifications allow for simple adjustments on an already existing cascading framework to further improve the resulting reconstructions.

Operator dependency of arterial input function in dynamic contrast-enhanced MRI.

OBJECTIVE: To investigate the effect of inter-operator variability in arterial input function (AIF) definition on kinetic parameter estimates (KPEs) from dynamic contrast-enhanced (DCE) MRI in patients with high-grade gliomas. METHODS: The study included 118 DCE series from 23 patients. AIFs were measured by three domain experts (DEs), and a population AIF (pop-AIF) was constructed from the measured AIFs. The DE-AIFs, pop-AIF and AUC-normalized DE-AIFs were used for pharmacokinetic analysis with the extended Tofts model. AIF-dependence of KPEs was assessed by intraclass correlation coefficient (ICC) analysis, and the impact on relative longitudinal change in Ktrans was assessed by Fleiss' kappa (κ). RESULTS: There was a moderate to substantial agreement (ICC 0.51-0.76) between KPEs when using DE-AIFs, while AUC-normalized AIFs yielded ICC 0.77-0.95 for Ktrans, kep and ve and ICC 0.70 for vp. Inclusion of the pop-AIF did not reduce agreement. Agreement in relative longitudinal change in Ktrans was moderate (κ = 0.591) using DE-AIFs, while AUC-normalized AIFs gave substantial (κ = 0.809) agreement. DISCUSSION: AUC-normalized AIFs can reduce the variation in kinetic parameter results originating from operator input. The pop-AIF presented in this work may be applied in absence of a satisfactory measurement.

Signal Enhancement of the Dentate Nucleus at Unenhanced MR Imaging after Very High Cumulative Doses of the Macrocyclic Gadolinium-based Contrast Agent Gadobutrol: An Observational Study.

Purpose To test for measurable visual enhancement of the dentate nucleus (DN) on unenhanced T1-weighted magnetic resonance (MR) images in a cohort of patients with a primary brain tumor who had not received linear gadolinium-based contrast agents (GBCAs) but had received many injections of macrocyclic GBCAs. Materials and Methods Seventeen patients with high-grade gliomas who had received 10-44 administrations of the macrocyclic GBCA gadobutrol (0.1 mmol/kg of body weight) were retrospectively included in this regional ethics committee-approved study. Two neuroradiologists inspected T1-weighted MR images with optimized window settings to visualize small differences in contrast at the baseline and at the last examination for the presence of visual DN signal enhancement. Signal intensity (SI) in the DN was normalized to the SI of the pons, and a one-sample t test was used to test for differences between baseline normalized SI (nSI) in the DN (nSIDN) and the average change in nSIDN of all postbaseline MR imaging sessions (ΔnSIDNavg) or the change in nSIDN from baseline to the last MR imaging session (ΔnSIDN). Linear and quadratic correlation analyses were used to examine the association between the number of macrocyclic GBCA administrations and ΔnSIDN or ΔnSIDNavg. Results The mean ± standard deviation number of macrocyclic GBCA administrations was 22.2 ± 10.6 administered throughout 706 days ± 454. Visually appreciable signal enhancement was observed in two patients who had received 37 and 44 macrocyclic GBCA injections. Mean ΔnSIDN was greater than zero (0.03 ± 0.05; P = .016), and there was a significant linear association between the number of macrocyclic GBCA injections and ΔnSIDN (r = 0.69, P = .002) and ΔnSIDNavg (r = 0.77, P < .001). Conclusion A small but statistically significant dose-dependent T1-weighted signal enhancement was observed in the DN after multiple macrocyclic GBCA injections. Visually appreciable enhancement in the DN was observed on contrast-optimized images in two patients who had received 37 and 44 standard doses of macrocyclic GBCAs. © RSNA, 2017 Online supplemental material is available for this article.

Voxel-Wise Perfusion Assessment in Cerebral White Matter with PCASL at 3T; Is It Possible and How Long Does It Take?

PURPOSE: To establish whether reliable voxel-wise assessment of perfusion in cerebral white matter (WM) is possible using arterial spin labeling (ASL) at 3T in a cohort of healthy subjects. MATERIALS AND METHODS: Pseudo-continuous ASL (PCASL) with background suppression (BS) optimized for WM measurements was performed at 3T in eight healthy male volunteers aged 25-41. Four different labeling schemes were evaluated by varying the labeling duration (LD) and post-labeling delay (PLD). Eight slices with voxel dimension 3.75x3.75x5 mm3 were acquired from the anterosuperior aspect of the brain, and 400 image/control pairs were collected for each run. Rigid head immobilization was applied using individually fitted thermoplastic masks. For each voxel in the resulting ASL time series, the time needed to reach a 95% significance level for the ASL signal to be higher than zero (paired t-test), was estimated. RESULTS: The four protocols detected between 88% and 95% (after Bonferroni correction: 75% and 88%) of WM voxels at 95% significance level. In the most efficient sequence, 80% was reached after 5 min and 95% after 53 min (after Bonferroni correction 40% and 88% respectively). For all protocols, the fraction of significant WM voxels increased in an asymptotic fashion with increasing scan time. A small subgroup of voxels was shown to not benefit at all from prolonged measurement. CONCLUSION: Acquisition of a significant ASL signal from a majority of WM voxels is possible within clinically acceptable scan times, whereas full coverage needs prohibitively long scan times, as a result of the asymptotic trajectory.

Correction of B0-distortions in echo-planar-imaging-based perfusion-weighted MRI.

PURPOSE: To evaluate and quantify a scheme for correcting susceptibility artifacts in spin-echo echo-planar-imaging-based dynamic susceptibility contrast (DSC) perfusion MRI of high-grade gliomas at 3 Tesla. MATERIALS AND METHODS: Sixteen patients with a total of 78 scans were studied. DSC-MRI images were corrected using a displacement map generated from opposite phase-encoding polarity images. Two methods were used for quantification in the correction: (i) linear regression of pixel-by-pixel comparisons, performed both globally and relative to the anterior and posterior commissure plane (AC-PC plane), of T2-weighted images with both corrected and uncorrected raw DSC images; and (ii) counting significant (>2.0) normalized cerebral blood volume (nCBV) pixels from perfusion maps in the tumor region of interest. RESULTS: Sixty-four of 78 datasets showed significant differences in the coefficient of correlation (r2) values. The difference between corrected and uncorrected r2 values was positive in all but one patient. Correction of B0- distortion significantly improved r2 in slices around the AC-PC plane. In 62% of the datasets, we observed an increased number of significant pixels in the corrected nCBV maps; 36% showed more significant pixels in uncorrected nCBV maps; 1% showed no difference. CONCLUSION: Distortion correction of DSC-MRI may provide improved accuracy compared with uncorrected data, especially for tumors located below the corpus callosum and near the frontal sinuses.