A multi-reader comparison of normal-appearing white matter normalization techniques for perfusion and diffusion MRI in brain tumors

Leveraging Physics-Based Synthetic MR Images and Deep Transfer Learning for Artifact Reduction in Echo-Planar Imaging

BACKGOUND AND PURPOSE

This study utilizes a physics-based approach to synthesize realistic MR artifacts and train a deep learning generative adversarial network (GAN) for use in artifact reduction on EPI, a crucial neuroimaging sequence with high acceleration that is notoriously susceptible to artifacts.

MATERIALS AND METHODS

A total of 4,573 anatomical MR sequences from 1,392 patients undergoing clinically indicated MRI of the brain were used to create a synthetic data set using physics-based, simulated artifacts commonly found in EPI. By using multiple MRI contrasts, we hypothesized the GAN would learn to correct common artifacts while preserving the inherent contrast information, even for contrasts the network has not been trained on. A modified Pix2PixGAN architecture with an Attention-R2UNet generator was used for the model. Three training strategies were employed: (1) An "all-in-one" model trained on all the artifacts at once; (2) a set of "single models", one for each artifact; and a (3) "stacked transfer learning" approach where a model is first trained on one artifact set, then this learning is transferred to a new model and the process is repeated for the next artifact set. Lastly, the "Stacked Transfer Learning" model was tested on ADC maps from single-shot diffusion MRI data in N = 49 patients diagnosed with recurrent glioblastoma to compare visual quality and lesion measurements between the natively acquired images and AI-corrected images.

RESULTS

The "stacked transfer learning" approach had superior artifact reduction performance compared to the other approaches as measured by Mean Squared Error (MSE = 0.0016), Structural Similarity Index (SSIM = 0.92), multiscale SSIM (MS-SSIM = 0.92), peak signal-to-noise ratio (PSNR = 28.10), and Hausdorff distance (HAUS = 4.08mm), suggesting that leveraging pre-trained knowledge and sequentially training on each artifact is the best approach this application. In recurrent glioblastoma, significantly higher visual quality was observed in model predicted images compared to native images, while quantitative measurements within the tumor regions remained consistent with non-corrected images.

CONCLUSIONS

The current study demonstrates the feasibility of using a physics-based method for synthesizing a large data set of images with realistic artifacts and the effectiveness of utilizing this synthetic data set in a "stacked transfer learning" approach to training a GAN for reduction of EPI-based artifacts.

"Synthetic" DSC Perfusion MRI with Adjustable Acquisition Parameters in Brain Tumors Using Dynamic Spin-and-Gradient-Echo Echoplanar Imaging

BACKGROUND AND PURPOSE

Normalized relative cerebral blood volume (nrCBV) and percentage of signal recovery (PSR) computed from dynamic susceptibility contrast (DSC) perfusion imaging are useful biomarkers for differential diagnosis and treatment response assessment in brain tumors. However, their measurements are dependent on DSC acquisition factors, and CBV-optimized protocols technically differ from PSR-optimized protocols. This study aimed to generate "synthetic" DSC data with adjustable synthetic acquisition parameters using dual-echo gradient-echo (GE) DSC datasets extracted from dynamic spin-and-gradient-echo echoplanar imaging (dynamic SAGE-EPI). Synthetic DSC was aimed at: 1) simultaneously create nrCBV and PSR maps using optimal sequence parameters, 2) compare DSC datasets with heterogeneous external cohorts, and 3) assess the impact of acquisition factors on DSC metrics.

MATERIALS AND METHODS

Thirty-eight patients with contrast-enhancing brain tumors were prospectively imaged with dynamic SAGE-EPI during a non-preloaded single-dose contrast injection and included in this cross-sectional study. Multiple synthetic DSC curves with desired pulse sequence parameters were generated using the Bloch equations applied to the dual-echo GE data extracted from dynamic SAGE-EPI datasets, with or without optional preload simulation.

RESULTS

Dynamic SAGE-EPI allowed for simultaneous generation of CBV-optimized and PSR-optimized DSC datasets with a single contrast injection, while PSR computation from guideline-compliant CBV-optimized protocols resulted in rank variations within the cohort (Spearman's ρ = 0.83-0.89, i.e. 31%-21% rank variation). Treatment-naïve glioblastoma exhibited lower parameter-matched PSR compared to the external cohorts of treatment-naïve primary CNS lymphomas (PCNSL) (p<0.0001), supporting a role of synthetic DSC for multicenter comparisons. Acquisition factors highly impacted PSR, and nrCBV without leakage correction also showed parameter-dependence, although less pronounced. However, this dependence was remarkably mitigated by post-hoc leakage correction.

CONCLUSIONS

Dynamic SAGE-EPI allows for simultaneous generation of CBV-optimized and PSR-optimized DSC data with one acquisition and a single contrast injection, facilitating the use of a single perfusion protocol for all DSC applications. This approach may also be useful for comparisons of perfusion metrics across heterogeneous multicenter datasets, as it facilitates post-hoc harmonization.

A review of deep learning for brain tumor analysis in MRI

Recent progress in deep learning (DL) is producing a new generation of tools across numerous clinical applications. Within the analysis of brain tumors in magnetic resonance imaging, DL finds applications in tumor segmentation, quantification, and classification. It facilitates objective and reproducible measurements crucial for diagnosis, treatment planning, and disease monitoring. Furthermore, it holds the potential to pave the way for personalized medicine through the prediction of tumor type, grade, genetic mutations, and patient survival outcomes. In this review, we explore the transformative potential of DL for brain tumor care and discuss existing applications, limitations, and future directions and opportunities.

Diffusion MRI is superior to quantitative T2-FLAIR mismatch in predicting molecular subtypes of human non-enhancing gliomas

PURPOSE

This study compared the classification performance of normalized apparent diffusion coefficient (nADC) with percentage T2-FLAIR mismatch-volume (%T2FM-volume) for differentiating between IDH-mutant astrocytoma (IDHm-A) and other glioma molecular subtypes.

METHODS

A total of 105 non-enhancing gliomas were studied. T2-FLAIR digital subtraction maps were used to identify T2FM and T2-FLAIR non-mismatch (T2FNM) subregions within tumor volumes of interest (VOIs). Median nADC from the whole tumor, T2FM, and T2NFM subregions and %T2FM-volume were obtained. IDHm-A classification analyses using receiver-operating characteristic curves and multiple logistic regression were performed in addition to exploratory survival analyses.

RESULTS

T2FM subregions had significantly higher nADC than T2FNM subregions within IDHm-A with ≥ 25% T2FM-volume (P < 0.0001). IDHm-A with ≥ 25% T2FM-volume demonstrated significantly higher whole tumor nADC compared to IDHm-A with < 25% T2FM-volume (P < 0.0001), and both IDHm-A subgroups demonstrated significantly higher nADC compared to IDH-mutant oligodendroglioma and IDH-wild-type gliomas (P < 0.05). For classification of IDHm-A vs. other gliomas, the area under curve (AUC) of nADC was significantly greater compared to the AUC of %T2FM-volume (P = 0.01, nADC AUC = 0.848, %T2FM-volume AUC = 0.714) along with greater sensitivity. In exploratory survival analyses within IDHm-A, %T2FM-volume was not associated with overall survival (P = 0.2), but there were non-significant trends for nADC (P = 0.07) and tumor volume (P = 0.051).

CONCLUSION

T2-FLAIR subtraction maps are useful for characterizing IDHm-A imaging characteristics. nADC outperforms %T2FM-volume for classifying IDHm-A amongst non-enhancing gliomas with preserved high specificity and increased sensitivity, which may be related to inherent diffusivity differences regardless of T2FM. In line with previous findings on visual T2FM-sign, quantitative %T2FM-volume may not be prognostic.

Differentiating IDH-mutant astrocytomas and 1p19q-codeleted oligodendrogliomas using DSC-PWI: high performance through cerebral blood volume and percentage of signal recovery percentiles

OBJECTIVE

Presurgical differentiation between astrocytomas and oligodendrogliomas remains an unresolved challenge in neuro-oncology. This research aims to provide a comprehensive understanding of each tumor's DSC-PWI signatures, evaluate the discriminative capacity of cerebral blood volume (CBV) and percentage of signal recovery (PSR) percentile values, and explore the synergy of CBV and PSR combination for pre-surgical differentiation.

METHODS

Patients diagnosed with grade 2 and 3 IDH-mutant astrocytomas and IDH-mutant 1p19q-codeleted oligodendrogliomas were retrospectively retrieved (2010-2022). 3D segmentations of each tumor were conducted, and voxel-level CBV and PSR were extracted to compute mean, minimum, maximum, and percentile values. Statistical comparisons were performed using the Mann-Whitney U test and the area under the receiver operating characteristic curve (AUC-ROC). Lastly, the five most discriminative variables were combined for classification with internal cross-validation.

RESULTS

The study enrolled 52 patients (mean age 45-year-old, 28 men): 28 astrocytomas and 24 oligodendrogliomas. Oligodendrogliomas exhibited higher CBV and lower PSR than astrocytomas across all metrics (e.g., mean CBV = 2.05 and 1.55, PSR = 0.68 and 0.81 respectively). The highest AUC-ROCs and the smallest p values originated from CBV and PSR percentiles (e.g., PSRp70 AUC-ROC = 0.84 and p value = 0.0005, CBVp75 AUC-ROC = 0.8 and p value = 0.0006). The mean, minimum, and maximum values yielded lower results. Combining the best five variables (PSRp65, CBVp70, PSRp60, CBVp75, and PSRp40) achieved a mean AUC-ROC of 0.87 for differentiation.

CONCLUSIONS

Oligodendrogliomas exhibit higher CBV and lower PSR than astrocytomas, traits that are emphasized when considering percentiles rather than mean or extreme values. The combination of CBV and PSR percentiles results in promising classification outcomes.

CLINICAL RELEVANCE STATEMENT

The combination of histogram-derived percentile values of cerebral blood volume and percentage of signal recovery from DSC-PWI enhances the presurgical differentiation between astrocytomas and oligodendrogliomas, suggesting that incorporating these metrics into clinical practice could be beneficial.

KEY POINTS

• The unsupervised selection of percentile values for cerebral blood volume and percentage of signal recovery enhances presurgical differentiation of astrocytomas and oligodendrogliomas. • Oligodendrogliomas exhibit higher cerebral blood volume and lower percentage of signal recovery than astrocytomas. • Cerebral blood volume and percentage of signal recovery combined provide a broader perspective on tumor vasculature and yield promising results for this preoperative classification.

Bright diffusion sign: A sensitive and specific radiologic biomarker for multinodular and vacuolating neuronal tumor

BACKGROUND AND PURPOSE

Accurate differentiation between multinodular and vacuolating neuronal tumor (MVNT) and dysembryoplastic neuroepithelial tumor (DNET) is important for treatment decision-making. We aimed to develop an accurate radiologic diagnostic model for differentiating MVNT from DNET using T2WI and diffusion-weighted imaging (DWI).

MATERIALS AND METHODS

A total of 56 patients (mean age, 47.48±17.78 years; 31 women) diagnosed with MVNT (n = 37) or DNET (n = 19) who underwent brain MRI, including T2WI and DWI, were included. Two board-certified neuroradiologists performed qualitative (bubble appearance, cortical involvement, bright diffusion sign, and bright apparent diffusion coefficient [ADC] sign) and quantitative (nDWI and nADC) assessments. A diagnostic tree model was developed with significant and reliable imaging findings using an exhaustive chi-squared Automatic Interaction Detector (CHAID) algorithm.

RESULTS

In visual assessment, the imaging features that showed high diagnostic accuracy and interobserver reliability were the bright diffusion sign and absence of cortical involvement (bright diffusion sign: accuracy, 94.64 %; sensitivity, 91.89 %; specificity, 100.00 %; interobserver agreement, 1.00; absence of cortical involvement: accuracy, 92.86 %; sensitivity, 89.19 %; specificity, 100.00 %; interobserver agreement, 1.00). In quantitative analysis, nDWI was significantly higher in MVNT than in DENT (1.52 ± 0.34 vs. 0.91 ± 0.27, p < 0.001), but the interobserver agreement was fair (intraclass correlation coefficient = 0.321). The overall diagnostic accuracy of the tree model with visual assessment parameters was 98.21 % (55/56).

CONCLUSION

The bright diffusion sign and absence of cortical involvement are accurate and reliable imaging findings for differentiating MVNT from DNET. By using simple, intuitive, and reliable imaging findings, such as the bright diffusion sign, MVNT can be accurately differentiated from DNET.

Quantification of T2-FLAIR Mismatch in Nonenhancing Diffuse Gliomas Using Digital Subtraction

BACKGROUND AND PURPOSE

The T2-FLAIR mismatch sign on MR imaging is a highly specific imaging biomarker of isocitrate dehydrogenase (IDH)-mutant astrocytomas, which lack 1p/19q codeletion. However, most studies using the T2-FLAIR mismatch sign have used visual assessment. This study quantified the degree of T2-FLAIR mismatch using digital subtraction of fluid-nulled T2-weighted FLAIR images from non-fluid-nulled T2-weighted images in human nonenhancing diffuse gliomas and then used this information to assess improvements in diagnostic performance and investigate subregion characteristics within these lesions.

MATERIALS AND METHODS

Two cohorts of treatment-naïve, nonenhancing gliomas with known IDH and 1p/19q status were studied (n = 71 from The Cancer Imaging Archive (TCIA) and n = 34 in the institutional cohort). 3D volumes of interest corresponding to the tumor were segmented, and digital subtraction maps of T2-weighted MR imaging minus T2-weighted FLAIR MR imaging were used to partition each volume of interest into a T2-FLAIR mismatched subregion (T2-FLAIR mismatch, corresponding to voxels with positive values on the subtraction maps) and nonmismatched subregion (T2-FLAIR nonmismatch corresponding to voxels with negative values on the subtraction maps). Tumor subregion volumes, percentage of T2-FLAIR mismatch volume, and T2-FLAIR nonmismatch subregion thickness were calculated, and 2 radiologists assessed the T2-FLAIR mismatch sign with and without the aid of T2-FLAIR subtraction maps.

RESULTS

Thresholds of ≥42% T2-FLAIR mismatch volume classified IDH-mutant astrocytoma with a specificity/sensitivity of 100%/19.6% (TCIA) and 100%/31.6% (institutional); ≥25% T2-FLAIR mismatch volume showed 92.0%/32.6% and 100%/63.2% specificity/sensitivity, and ≥15% T2-FLAIR mismatch volume showed 88.0%/39.1% and 93.3%/79.0% specificity/sensitivity. In IDH-mutant astrocytomas with ≥15% T2-FLAIR mismatch volume, T2-FLAIR nonmismatch subregion thickness was negatively correlated with the percentage T2-FLAIR mismatch volume (P < .0001) across both cohorts. The percentage T2-FLAIR mismatch volume was higher in grades 3-4 compared with grade 2 IDH-mutant astrocytomas (P < .05), and ≥15% T2-FLAIR mismatch volume IDH-mutant astrocytomas were significantly larger than <15% T2-FLAIR mismatch volume IDH-mutant astrocytoma (P < .05) across both cohorts. When evaluated by 2 radiologists, the additional use of T2-FLAIR subtraction maps did not show a significant difference in interreader agreement, sensitivity, or specificity compared with a separate evaluation of T2-FLAIR and T2-weighted MR imaging alone.

CONCLUSIONS

T2-FLAIR digital subtraction maps may be a useful, automated tool to obtain objective segmentations of tumor subregions based on quantitative thresholds for classifying IDH-mutant astrocytomas using the percentage T2 FLAIR mismatch volume with 100% specificity and exploring T2-FLAIR mismatch/T2-FLAIR nonmismatch subregion characteristics. Conversely, the addition of T2-FLAIR subtraction maps did not enhance the sensitivity or specificity of the visual T2-FLAIR mismatch sign assessment by experienced radiologists.

Diffusion histogram profiles predict molecular features of grade 4 in histologically lower-grade adult diffuse gliomas following WHO classification 2021

OBJECTIVES

In the latest World Health Organization classification 2021, grade 4 adult diffuse gliomas can be diagnosed with several molecular features even without histological evidence of necrosis or microvascular proliferation. We aimed to explore whole tumor histogram-derived apparent diffusion coefficient (ADC) histogram profiles for differentiating between the presence (Mol-4) and absence (Mol-2/3) of grade 4 molecular features in histologically lower-grade gliomas.

METHODS

Between June 2019 and October 2022, 184 adult patients with diffuse gliomas underwent MRI. After excluding 121 patients, 18 (median age, 64.5 [range, 37-84 years]) Mol-4 and 45 (median 40 [range, 18-73] years) Mol-2/3 patients with histologically lower-grade gliomas were enrolled. Whole tumor volume-of-interest-derived ADC histogram profiles were calculated and compared between the two groups. Stepwise logistic regression analysis with Akaike's information criterion using the ADC histogram profiles with p values < 0.01 and age at diagnosis was used to identify independent variables for predicting the Mol-4 group.

RESULTS

The 90th percentile (p < 0.001), median (p < 0.001), mean (p < 0.001), 10th percentile (p = 0.014), and entropy (p < 0.001) of normalized ADC were lower, and kurtosis (p < 0.001) and skewness (p = 0.046) were higher in the Mol-4 group than in the Mol-2/3 group. Multivariate logistic regression analysis revealed that the entropy of normalized ADC and age at diagnosis were independent predictive parameters for the Mol-4 group with an area under the curve of 0.92.

CONCLUSION

ADC histogram profiles may be promising preoperative imaging biomarkers to predict molecular grade 4 among histologically lower-grade adult diffuse gliomas.

CLINICAL RELEVANCE STATEMENT

This study highlighted the diagnostic usefulness of ADC histogram profiles to differentiate histologically lower grade adult diffuse gliomas with the presence of molecular grade 4 features and those without.

KEY POINTS

• ADC histogram profiles to predict molecular CNS WHO grade 4 status among histologically lower-grade adult diffuse gliomas were evaluated. • Entropy of ADC and age were independent predictive parameters for molecular grade 4 status. • ADC histogram analysis is useful for predicting molecular grade 4 among histologically lower-grade gliomas.

Multi-nuclear sodium, diffusion, and perfusion MRI in human gliomas

PURPOSE

There is limited knowledge about the associations between sodium and proton MRI measurements in brain tumors. The purpose of this study was to quantify intra- and intertumoral correlations between sodium, diffusion, and perfusion MRI in human gliomas.

METHODS

Twenty glioma patients were prospectively studied on a 3T MRI system with multinuclear capabilities. Three mutually exclusive tumor volumes of interest (VOIs) were segmented: contrast-enhancing tumor (CET), T2/FLAIR hyperintense non-enhancing tumor (NET), and necrosis. Median and voxel-wise associations between apparent diffusion coefficient (ADC), normalized relative cerebral blood volume (nrCBV), and normalized sodium measurements were quantified for each VOI.

RESULTS

Both relative sodium concentration and ADC were significantly higher in areas of necrosis compared to NET (P = 0.003 and P = 0.008, respectively) and CET (P = 0.02 and P = 0.02). Sodium concentration was higher in CET compared to NET (P = 0.04). Sodium and ADC were higher in treated compared to treatment-naïve gliomas within NET (P = 0.006 and P = 0.01, respectively), and ADC was elevated in CET (P = 0.03). Median ADC and sodium concentration were positively correlated across patients in NET (r = 0.77, P < 0.0001) and CET (r = 0.84, P < 0.0001), but not in areas of necrosis (r = 0.45, P = 0.12). Median nrCBV and sodium concentration were negatively correlated across patients in areas of NET (r=-0.63, P = 0.003). Similar associations were observed when examining voxel-wise correlations within VOIs.

CONCLUSION

Sodium MRI is positively correlated with proton diffusion MRI measurements in gliomas, likely reflecting extracellular water. Unique areas of multinuclear MRI contrast may be useful in future studies to understand the chemistry of the tumor microenvironment.