Automated brain extraction of multisequence MRI using artificial neural networks

Semi-Supervised Learning in Medical MRI Segmentation: Brain Tissue with White Matter Hyperintensity Segmentation Using FLAIR MRI

White-matter hyperintensity (WMH) is a primary biomarker for small-vessel cerebrovascular disease, Alzheimer's disease (AD), and others. The association of WMH with brain structural changes has also recently been reported. Although fluid-attenuated inversion recovery (FLAIR) magnetic resonance imaging (MRI) provide valuable information about WMH, FLAIR does not provide other normal tissue information. The multi-modal analysis of FLAIR and T1-weighted (T1w) MRI is thus desirable for WMH-related brain aging studies. In clinical settings, however, FLAIR is often the only available modality. In this study, we thus propose a semi-supervised learning method for full brain segmentation using FLAIR. The results of our proposed method were compared with the reference labels, which were obtained by FreeSurfer segmentation on T1w MRI. The relative volume difference between the two sets of results shows that our proposed method has high reliability. We further evaluated our proposed WMH segmentation by comparing the Dice similarity coefficients of the reference and the results of our proposed method. We believe our semi-supervised learning method has a great potential for use for other MRI sequences and will encourage others to perform brain tissue segmentation using MRI modalities other than T1w.

Lesion-Function Analysis from Multimodal Imaging and Normative Brain Atlases for Prediction of Cognitive Deficits in Glioma Patients

Simple Summary

This prospective cross-sectional study utilized standard structural MR imaging and amino acid PET in conjunction with brain atlases of gray matter functional regions and white matter tracts, and elastic registration techniques to estimate the influence of the type and location of treatment-related brain damage or recurrent tumors on cognitive functioning in a group of well-doing WHO Grade III/IV glioma patients at follow-up after treatment. The negative impact of T2/FLAIR hyperintensities, supposed to be mainly caused by radiotherapy, on cognitive performance far exceeded that of surgical brain defects or recurrent tumors. The affection of functional nodes and fiber tracts of the left hemisphere and especially of the left temporal lobe by T2/FLAIR hyperintensities was highly correlated with verbal episodic memory dysfunction. These observations imply that radiotherapy for gliomas of the left hemisphere should be individually tailored by means of publicly available brain atlases and registration techniques.

Abstract

Cognitive deficits are common in glioma patients following multimodality therapy, but the relative impact of different types and locations of treatment-related brain damage and recurrent tumors on cognition is not well understood. In 121 WHO Grade III/IV glioma patients, structural MRI, O-(2-[18F]fluoroethyl)-L-tyrosine FET-PET, and neuropsychological testing were performed at a median interval of 14 months (range, 1–214 months) after therapy initiation. Resection cavities, T1-enhancing lesions, T2/FLAIR hyperintensities, and FET-PET positive tumor sites were semi-automatically segmented and elastically registered to a normative, resting state (RS) fMRI-based functional cortical network atlas and to the JHU atlas of white matter (WM) tracts, and their influence on cognitive test scores relative to a cohort of matched healthy subjects was assessed. T2/FLAIR hyperintensities presumably caused by radiation therapy covered more extensive brain areas than the other lesion types and significantly impaired cognitive performance in many domains when affecting left-hemispheric RS-nodes and WM-tracts as opposed to brain tissue damage caused by resection or recurrent tumors. Verbal episodic memory proved to be especially vulnerable to T2/FLAIR abnormalities affecting the nodes and tracts of the left temporal lobe. In order to improve radiotherapy planning, publicly available brain atlases, in conjunction with elastic registration techniques, should be used, similar to neuronavigation in neurosurgery.

Generative adversarial network for glioblastoma ensures morphologic variations and improves diagnostic model for isocitrate dehydrogenase mutant type

Generative adversarial network (GAN) creates synthetic images to increase data quantity, but whether GAN ensures meaningful morphologic variations is still unknown. We investigated whether GAN-based synthetic images provide sufficient morphologic variations to improve molecular-based prediction, as a rare disease of isocitrate dehydrogenase (IDH)-mutant glioblastomas. GAN was initially trained on 500 normal brains and 110 IDH-mutant high-grade astocytomas, and paired contrast-enhanced T1-weighted and FLAIR MRI data were generated. Diagnostic models were developed from real IDH-wild type (n = 80) with real IDH-mutant glioblastomas (n = 38), or with synthetic IDH-mutant glioblastomas, or augmented by adding both real and synthetic IDH-mutant glioblastomas. Turing tests showed synthetic data showed reality (classification rate of 55%). Both the real and synthetic data showed that a more frontal or insular location (odds ratio [OR] 1.34 vs. 1.52; P = 0.04) and distinct non-enhancing tumor margins (OR 2.68 vs. 3.88; P < 0.001), which become significant predictors of IDH-mutation. In an independent validation set, diagnostic accuracy was higher for the augmented model (90.9% [40/44] and 93.2% [41/44] for each reader, respectively) than for the real model (84.1% [37/44] and 86.4% [38/44] for each reader, respectively). The GAN-based synthetic images yield morphologically variable, realistic-seeming IDH-mutant glioblastomas. GAN will be useful to create a realistic training set in terms of morphologic variations and quality, thereby improving diagnostic performance in a clinical model.

Detection of Isocitrate Dehydrogenase Mutated Glioblastomas Through Anomaly Detection Analytics

BACKGROUND

The rarity of Isocitrate Dehydrogenase mutated (mIDH) glioblastomas relative to wild-type IDH glioblastomas, as well as their distinct tumor physiology, effectively render them "outliers". Specialized tools are needed to identify these outliers.

OBJECTIVE

To carefully craft and apply anomaly detection methods to identify mIDH glioblastoma based on radiomic features derived from magnetic resonance imaging.

METHODS

T1-post gadolinium images for 188 patients and 138 patients were downloaded from The Cancer Imaging Archive's (TCIA) The Cancer Genome Atlas (TCGA) glioblastoma collection, and from the University of Minnesota Medical Center (UMMC), respectively. Anomaly detection methods were tested on glioblastoma image features for the precision of mIDH detection and compared to standard classification methods.

RESULTS

Using anomaly detection training methods, we were able to detect IDH mutations from features in noncontrast-enhancing regions in glioblastoma with an average precision of 75.0%, 69.9%, and 69.8% using three different models. Anomaly detection methods consistently outperformed traditional two-class classification methods from 2 unique learning models (67.9%, 67.6%). The disparity in performances could not be overcome through newer, popular models such as neural networks (67.4%).

CONCLUSION

We employed an anomaly detection strategy in the detection of IDH mutation in glioblastoma using preoperative T1 postcontrast imaging. We show these methods outperform traditional two-class classification in the setting of dataset imbalances inherent to IDH mutation prevalence in glioblastoma. We validate our results using an external dataset and highlight new possible avenues for radiogenomic rare event prediction in glioblastoma and beyond.

Age- and Intravenous Methotrexate-Associated Leukoencephalopathy and Its Neurological Impact in Pediatric Patients with Lymphoblastic Leukemia

Simple Summary

In this study, we investigated standardized post-chemotherapy magnetic resonance (MR) scans for leukoencephalopathy and patient- and treatment-related risk factors in childhood leukemia patients. As prevalence numbers are limited, our study provides the required estimations for this population. Furthermore, we demonstrate that younger patients might be more at-risk for development of leukoencephalopathy (LE), and that a higher intravenous methotrexate (IV-MTX) dose has a cumulative toxic effect, while the number of intrathecal administrations was not significantly associated with the extent of LE. This can suggest we should modify chemotherapeutic treatment regimens by decreasing the number of IV-MTX applications, with special attention for younger patients.

Abstract

Methotrexate (MTX) is associated with leukoencephalopathy (LE) in children treated for lymphoblastic leukemia/lymphoma (ALL/LBL). However, large-scale studies with systematic MR acquisition and quantitative volumetric lesion information remain limited. Hence, the prevalence of lesion burdens and the potential risk factors of LE in this population are still inconclusive. FLAIR-MRI scans were acquired at the end of treatment in children who were treated for ALL/LBL, which were quantitatively analyzed for LE. Voxels were assigned to the lesion segmentation if indicated by two raters. Logistic and linear regression models were used to test whether lesion presence and size were predicted by risk factors such as age at diagnosis, gender, intrathecal (IT-) or intravenous (IV-)MTX dose, CNS invasion, and acute neurological events. Patients with a pre-existing neurological condition or low-quality MR scan were excluded from the analyses. Of the 129 patients, ten (8%) suffered from CNS invasion. Chemotherapy-associated neurological events were observed in 13 patients (10%) during therapy, and 68 patients (53%) showed LE post-treatment. LE was more frequent in cases of lower age and higher cumulative IV-MTX doses, while the extent of LE and neurological symptoms were associated only with IV-MTX doses. Neurological events were not significantly associated with LE, even though symptomatic patients demonstrated a higher ratio of LE (n = 9/13) than asymptomatic patients (n = 59/116). This study suggests leukoencephalopathy frequently occurs in both symptomatic and asymptomatic leukemia patients. Younger children and patients treated with higher cumulative IV-MTX doses might need more regular screening for early detection and follow-up of associated sequelae.

Diagnostic biomarkers from proteomic characterization of cerebrospinal fluid in patients with brain malignancies

Recent technological advances in molecular diagnostics through liquid biopsies hold the promise to repetitively monitor tumor evolution and treatment response of brain malignancies without the need of invasive surgical tissue accrual. Here, we implemented a mass spectrometry-based protein analysis pipeline which identified hundreds of proteins in 251 cerebrospinal fluid (CSF) samples from patients with four types of brain malignancies (glioblastoma, lymphoma, brain metastasis, and leptomeningeal disease [LMD]) and from healthy individuals with a focus on glioblastoma in a retrospective and confirmatory prospective observational study. CSF proteome deregulation via disruption of the blood brain barrier appeared to be largely conserved across brain tumor entities. CSF analysis of glioblastoma patients identified two proteomic clusters that correlated with tumor size and patient survival. By integrating CSF data with proteomic analyses of matching glioblastoma tumor tissue and primary glioblastoma cells, we identified potential CSF biomarkers for glioblastoma, in particular chitinase-3-like protein 1 (CHI3L1) and glial fibrillary acidic protein (GFAP). Key findings were validated in a prospective cohort consisting of 35 glioma patients. Finally, in LMD patients who frequently undergo repeated CSF work-up, we explored our proteomic pipeline as a mean to profile consecutive CSF samples. Therefore, proteomic analysis of CSF in brain malignancies has the potential to reveal biomarkers for diagnosis and therapy monitoring.

Effect of Applying Leakage Correction on rCBV Measurement Derived From DSC-MRI in Enhancing and Nonenhancing Glioma

Purpose

Relative cerebral blood volume (rCBV) is the most widely used parameter derived from DSC perfusion MR imaging for predicting brain tumor aggressiveness. However, accurate rCBV estimation is challenging in enhancing glioma, because of contrast agent extravasation through a disrupted blood-brain barrier (BBB), and even for nonenhancing glioma with an intact BBB, due to an elevated steady-state contrast agent concentration in the vasculature after first passage. In this study a thorough investigation of the effects of two different leakage correction algorithms on rCBV estimation for enhancing and nonenhancing tumors was conducted.

Methods

Two datasets were used retrospectively in this study: 1. A publicly available TCIA dataset (49 patients with 35 enhancing and 14 nonenhancing glioma); 2. A dataset acquired clinically at Erasmus MC (EMC, Rotterdam, NL) (47 patients with 20 enhancing and 27 nonenhancing glial brain lesions). The leakage correction algorithms investigated in this study were: a unidirectional model-based algorithm with flux of contrast agent from the intra- to the extravascular extracellular space (EES); and a bidirectional model-based algorithm additionally including flow from EES to the intravascular space.

Results

In enhancing glioma, the estimated average contrast-enhanced tumor rCBV significantly (Bonferroni corrected Wilcoxon Signed Rank Test, p < 0.05) decreased across the patients when applying unidirectional and bidirectional correction: 4.00 ± 2.11 (uncorrected), 3.19 ± 1.65 (unidirectional), and 2.91 ± 1.55 (bidirectional) in TCIA dataset and 2.51 ± 1.3 (uncorrected), 1.72 ± 0.84 (unidirectional), and 1.59 ± 0.9 (bidirectional) in EMC dataset. In nonenhancing glioma, a significant but smaller difference in observed rCBV was found after application of both correction methods used in this study: 1.42 ± 0.60 (uncorrected), 1.28 ± 0.46 (unidirectional), and 1.24 ± 0.37 (bidirectional) in TCIA dataset and 0.91 ± 0.49 (uncorrected), 0.77 ± 0.37 (unidirectional), and 0.67 ± 0.34 (bidirectional) in EMC dataset.

Conclusion

Both leakage correction algorithms were found to change rCBV estimation with BBB disruption in enhancing glioma, and to a lesser degree in nonenhancing glioma. Stronger effects were found for bidirectional leakage correction than for unidirectional leakage correction.

Brain tissues have single-voxel signatures in multi-spectral MRI

Since the seminal works by Brodmann and contemporaries, it is well-known that different brain regions exhibit unique cytoarchitectonic and myeloarchitectonic features. Transferring the approach of classifying brain tissues - and other tissues - based on their intrinsic features to the realm of magnetic resonance (MR) is a longstanding endeavor. In the 1990s, atlas-based segmentation replaced earlier multi-spectral classification approaches because of the large overlap between the class distributions. Here, we explored the feasibility of performing global brain classification based on intrinsic MR features, and used several technological advances: ultra-high field MRI, q-space trajectory diffusion imaging revealing voxel-intrinsic diffusion properties, chemical exchange saturation transfer and semi-solid magnetization transfer imaging as a marker of myelination and neurochemistry, and current neural network architectures to analyze the data. In particular, we used the raw image data as well to increase the number of input features. We found that a global brain classification of roughly 97 brain regions was feasible with gross classification accuracy of 60%; and that mapping from voxel-intrinsic MR data to the brain region to which the data belongs is possible. This indicates the presence of unique MR signals of different brain regions, similar to their cytoarchitectonic and myeloarchitectonic fingerprints.

Analyzing magnetic resonance imaging data from glioma patients using deep learning

The quantitative analysis of images acquired in the diagnosis and treatment of patients with brain tumors has seen a significant rise in the clinical use of computational tools. The underlying technology to the vast majority of these tools are machine learning methods and, in particular, deep learning algorithms. This review offers clinical background information of key diagnostic biomarkers in the diagnosis of glioma, the most common primary brain tumor. It offers an overview of publicly available resources and datasets for developing new computational tools and image biomarkers, with emphasis on those related to the Multimodal Brain Tumor Segmentation (BraTS) Challenge. We further offer an overview of the state-of-the-art methods in glioma image segmentation, again with an emphasis on publicly available tools and deep learning algorithms that emerged in the context of the BraTS challenge.

The traveling heads 2.0: Multicenter reproducibility of quantitative imaging methods at 7 Tesla

OBJECT

This study evaluates inter-site and intra-site reproducibility at ten different 7 T sites for quantitative brain imaging.

MATERIAL AND METHODS

Two subjects - termed the "traveling heads" - were imaged at ten different 7 T sites with a harmonized quantitative brain MR imaging protocol. In conjunction with the system calibration, MP2RAGE, QSM, CEST and multi-parametric mapping/relaxometry were examined.

RESULTS

Quantitative measurements with MP2RAGE showed very high reproducibility across sites and subjects, and errors were in concordance with previous results and other field strengths. QSM had high inter-site reproducibility for relevant subcortical volumes. CEST imaging revealed systematic differences between the sites, but reproducibility was comparable to results in the literature. Relaxometry had also very high agreement between sites, but due to the high sensitivity, differences caused by different applications of the B1 calibration of the two RF coil types used were observed.

CONCLUSION

Our results show that quantitative brain imaging can be performed with high reproducibility at 7 T and with similar reliability as found at 3 T for multicenter studies of the supratentorial brain.

Spatiotemporal habitats from multiparametric physiologic MRI distinguish tumor progression from treatment-related change in post-treatment glioblastoma

OBJECTIVES

We aimed to develop multiparametric physiologic MRI-based spatial habitats and to evaluate whether temporal changes in these habitats help to distinguish tumor progression from treatment-related change in post-treatment glioblastoma.

METHODS

This retrospective, single-institution study included patients with glioblastoma treated by concurrent chemoradiotherapy who had newly developed or enlarging, measurable contrast-enhancing mass. Contrast-enhancing mass was divided into three spatial habitats by K-means clustering of voxel-wise ADC and CBV values. Temporal changes of these habitats between two consecutive examinations prior to the diagnosis of tumor progression or treatment-related change were assessed. Predictors were selected using logistic regression and the performance was measured with an area under the receiver operating characteristics curve (AUC). Spatiotemporal habitats were further analyzed for correlation with the site of tumor progression.

RESULTS

There were 75 patients (mean, 58 years; range, 26-81 years; 43 men) with 48 cases of tumor progression and 39 cases of treatment-related change including 12 patient overlaps at different time points. Three spatial habitats of hypervascular cellular, hypovascular cellular, and nonviable tissue were identified. Increase in the hypervascular cellular (OR 4.55, p = .002) and hypovascular cellular habitat (OR 1.22, p < .001) was predictive of tumor progression. Combination of spatiotemporal habitats yielded a high diagnostic performance with an AUC of 0.89 (95% CI, 0.87-0.92). An increase in hypovascular cellular habitat predicted the site of tumor progression in 84% [21/25] of cases with tumor progression.

CONCLUSIONS

Temporal changes in spatial habitats derived from multiparametric physiologic MRI provided diagnostic value in distinguishing tumor progression from treatment-related change and predicted site of tumor progression in post-treatment glioblastoma.

KEY POINTS

• In post-treatment glioblastoma, three spatial habitats of hypervascular cellular, hypovascular cellular, and nonviable tissue were identified, and an increase in the hypervascular cellular (OR 4.55, p = .002) and hypovascular cellular habitat (OR 1.22, p < .001) was predictive of tumor progression. • Combination of spatiotemporal habitats yielded a high diagnostic performance with an AUC of 0.89 (95% CI, 0.87-0.92). • An increase in hypovascular cellular habitat predicted the site of tumor progression in 84% (21/25) of cases with tumor progression.

Virtual brain grafting: Enabling whole brain parcellation in the presence of large lesions

Brain atlases and templates are at the heart of neuroimaging analyses, for which they facilitate multimodal registration, enable group comparisons and provide anatomical reference. However, as atlas-based approaches rely on correspondence mapping between images they perform poorly in the presence of structural pathology. Whilst several strategies exist to overcome this problem, their performance is often dependent on the type, size and homogeneity of any lesions present. We therefore propose a new solution, referred to as Virtual Brain Grafting (VBG), which is a fully-automated, open-source workflow to reliably parcellate magnetic resonance imaging (MRI) datasets in the presence of a broad spectrum of focal brain pathologies, including large, bilateral, intra- and extra-axial, heterogeneous lesions with and without mass effect. The core of the VBG approach is the generation of a lesion-free T1-weighted image, which enables further image processing operations that would otherwise fail. Here we validated our solution based on Freesurfer recon-all parcellation in a group of 10 patients with heterogeneous gliomatous lesions, and a realistic synthetic cohort of glioma patients (n = 100) derived from healthy control data and patient data. We demonstrate that VBG outperforms a non-VBG approach assessed qualitatively by expert neuroradiologists and Mann-Whitney U tests to compare corresponding parcellations (real patients U(6,6) = 33, z = 2.738, P < .010, synthetic-patients U(48,48) = 2076, z = 7.336, P < .001). Results were also quantitatively evaluated by comparing mean dice scores from the synthetic-patients using one-way ANOVA (unilateral VBG = 0.894, bilateral VBG = 0.903, and non-VBG = 0.617, P < .001). Additionally, we used linear regression to show the influence of lesion volume, lesion overlap with, and distance from the Freesurfer volumes of interest, on labeling accuracy. VBG may benefit the neuroimaging community by enabling automated state-of-the-art MRI analyses in clinical populations using methods such as FreeSurfer, CAT12, SPM, Connectome Workbench, as well as structural and functional connectomics. To fully maximize its availability, VBG is provided as open software under a Mozilla 2.0 license (https://github.com/KUL-Radneuron/KUL_VBG).

Deep Learning-Based Method to Differentiate Neuromyelitis Optica Spectrum Disorder From Multiple Sclerosis

Background: Differentiating neuromyelitis optica spectrum disorder (NMOSD) from multiple sclerosis (MS) is crucial in the field of diagnostics because, despite their similarities, the treatments for these two diseases are substantially different, and disease-modifying treatments for MS can worsen NMOSD. As brain magnetic resonance imaging (MRI) is an important tool to distinguish the two diseases, extensive research has been conducted to identify the defining characteristics of MRI images corresponding to these two diseases. However, the application of such research in clinical practice is still limited. In this study, we investigate the applicability of a deep learning-based algorithm for differentiating NMOSD from MS.

Methods: In this study, we included 338 participants (213 patients with MS, 125 patients with NMOSD) who visited the Asan medical center between February 2009 and February 2020. A 3D convolutional neural network, which is a deep learning-based algorithm, was trained using fluid-attenuated inversion recovery images and clinical information of the participants. The performance of the final model in differentiating NMOSD from MS was evaluated and compared with that of two neurologists.

Results: The deep learning-based model exhibited an area under the receiver operating characteristic curve of 0.82 (95% CI, 0.75–0.89). It differentiated NMOSD from MS with an accuracy of 71.1% (sensitivity = 87.8%, specificity = 61.6%), which is comparable to that exhibited by the neurologists. The intra-rater reliability of the two neurologists was moderate (κ = 0.47, 0.50), which was in contrast with the consistent classification of the deep learning-based model.

Conclusion: The proposed model was verified to be capable of differentiating NMOSD from MS with accuracy comparable to that of neurologists, exhibiting the advantage of consistent classification. As a result, it can aid differential diagnosis between two important central nervous system inflammatory diseases in clinical practice.

Brain extraction on MRI scans in presence of diffuse glioma: Multi-institutional performance evaluation of deep learning methods and robust modality-agnostic training

Brain extraction, or skull-stripping, is an essential pre-processing step in neuro-imaging that has a direct impact on the quality of all subsequent processing and analyses steps. It is also a key requirement in multi-institutional collaborations to comply with privacy-preserving regulations. Existing automated methods, including Deep Learning (DL) based methods that have obtained state-of-the-art results in recent years, have primarily targeted brain extraction without considering pathologically-affected brains. Accordingly, they perform sub-optimally when applied on magnetic resonance imaging (MRI) brain scans with apparent pathologies such as brain tumors. Furthermore, existing methods focus on using only T1-weighted MRI scans, even though multi-parametric MRI (mpMRI) scans are routinely acquired for patients with suspected brain tumors. In this study, we present a comprehensive performance evaluation of recent deep learning architectures for brain extraction, training models on mpMRI scans of pathologically-affected brains, with a particular focus on seeking a practically-applicable, low computational footprint approach, generalizable across multiple institutions, further facilitating collaborations. We identified a large retrospective multi-institutional dataset of n=3340 mpMRI brain tumor scans, with manually-inspected and approved gold-standard segmentations, acquired during standard clinical practice under varying acquisition protocols, both from private institutional data and public (TCIA) collections. To facilitate optimal utilization of rich mpMRI data, we further introduce and evaluate a novel ''modality-agnostic training'' technique that can be applied using any available modality, without need for model retraining. Our results indicate that the modality-agnostic approach1 obtains accurate results, providing a generic and practical tool for brain extraction on scans with brain tumors.

Spatiotemporal Heterogeneity in Multiparametric Physiologic MRI Is Associated with Patient Outcomes in IDH-Wildtype Glioblastoma

PURPOSE

Heterogeneity in glioblastomas is associated with poorer outcomes, and physiologic heterogeneity can be quantified with noninvasive imaging. We developed spatial habitats based on multiparametric physiologic MRI and evaluated associations between temporal changes in these habitats and progression-free survival (PFS) after concurrent chemoradiotherapy (CCRT) in patients with glioblastoma.

EXPERIMENTAL DESIGN

Ninety-seven patients with isocitrate dehydrogenase (IDH)-wildtype glioblastoma were enrolled and two serial MRI examinations after CCRT were analyzed. Cerebral blood volumes and apparent diffusion coefficients were grouped using k-means clustering into three spatial habitats. Associations between temporal changes in spatial habitats and PFS were investigated using Cox proportional hazard modeling. The performance of significant predictors for PFS and overall survival (OS) was measured using a discrete increase of habitat (habitat risk score) in a temporal validation set from a prospective registry (n = 53, ClinicalTrials.gov NCT02619890). The site of progression was matched with the spatiotemporal habitats.

RESULTS

Three spatial habitats of hypervascular cellular, hypovascular cellular, and nonviable tissue were identified. A short-term increase in the hypervascular cellular habitat (HR, 40.0; P = 0.001) and hypovascular cellular habitat was significantly associated with shorter PFS (HR, 3.78; P < 0.001) after CCRT. Combined with clinical predictors, the habitat risk score showed a C-index of 0.79 for PFS and 0.74 for OS and stratified patients with short, intermediate, and long PFS (P = 0.016). An increase in the hypovascular cellular habitat predicted tumor progression sites.

CONCLUSIONS

Hypovascular cellular habitats derived from multiparametric physiologic MRIs may be useful predictors of clinical outcomes in patients with posttreatment glioblastoma.

Deep learning of brain magnetic resonance images: A brief review

Magnetic resonance imaging (MRI) is one of the most popular techniques in brain science and is important for understanding brain function and neuropsychiatric disorders. However, the processing and analysis of MRI is not a trivial task with lots of challenges. Recently, deep learning has shown superior performance over traditional machine learning approaches in image analysis. In this survey, we give a brief review of the recent popular deep learning approaches and their applications in brain MRI analysis. Furthermore, popular brain MRI databases and deep learning tools are also introduced. The strength and weaknesses of different approaches are addressed, and challenges as well as future directions are also discussed.

Noninvasive Characterization of Tumor Angiogenesis and Oxygenation in Bevacizumab-treated Recurrent Glioblastoma by Using Dynamic Susceptibility MRI: Secondary Analysis of the European Organization for Research and Treatment of Cancer 26101 Trial

Background Relevance of antiangiogenic treatment with bevacizumab in patients with glioblastoma is controversial because progression-free survival benefit did not translate into an overall survival (OS) benefit in randomized phase III trials. Purpose To perform longitudinal characterization of intratumoral angiogenesis and oxygenation by using dynamic susceptibility contrast agent-enhanced (DSC) MRI and evaluate its potential for predicting outcome from administration of bevacizumab. Materials and Methods In this secondary analysis of the prospective randomized phase II/III European Organization for Research and Treatment of Cancer 26101 trial conducted between October 2011 and December 2015 in 596 patients with first recurrence of glioblastoma, the subset of patients with availability of anatomic MRI and DSC MRI at baseline and first follow-up was analyzed. Patients were allocated into those administered bevacizumab (hereafter, the BEV group; either bevacizumab monotherapy or bevacizumab with lomustine) and those not administered bevacizumab (hereafter, the non-BEV group with lomustine monotherapy). Contrast-enhanced tumor volume, noncontrast-enhanced T2 fluid-attenuated inversion recovery (FLAIR) signal abnormality volume, Gaussian-normalized relative cerebral blood volume (nrCBV), Gaussian-normalized relative blood flow (nrCBF), and tumor metabolic rate of oxygen (nTMRO₂) was quantified. The predictive ability of these imaging parameters was assessed with multivariable Cox regression and formal interaction testing. Results A total of 254 of 596 patients were evaluated (mean age, 57 years ± 11; 155 men; 161 in the BEV group and 93 in non-BEV group). Progression-free survival was longer in the BEV group (3.7 months; 95% confidence interval [CI]: 3.0, 4.2) compared with the non-BEV group (2.5 months; 95% CI: 1.5, 2.9; P = .01), whereas OS was not different (P = .15). The nrCBV decreased for the BEV group (-16.3%; interquartile range [IQR], -39.5% to 12.0%; P = .01), but not for the non-BEV group (1.2%; IQR, -17.9% to 23.3%; P = .19) between baseline and first follow-up. An identical pattern was observed for both nrCBF and nTMRO₂ values. Contrast-enhanced tumor and noncontrast-enhanced T2 FLAIR signal abnormality volumes decreased for the BEV group (-66% [IQR, -83% to -35%] and -33% [IQR, -71% to -5%], respectively; P < .001 for both), whereas they increased for the non-BEV group (30% [IQR, -17% to 98%], P = .001; and 10% [IQR, -13% to 82%], P = .02, respectively) between baseline and first follow-up. None of the assessed MRI parameters were predictive for OS in the BEV group. Conclusion Bevacizumab treatment decreased tumor volumes, angiogenesis, and oxygenation, thereby reflecting its effectiveness for extending progression-free survival; however, these parameters were not predictive of overall survival (OS), which highlighted the challenges of identifying patients that derive an OS benefit from bevacizumab. © RSNA, 2020 Online supplemental material is available for this article. See also the editorial by Dillon in this issue.

BraTS Toolkit: Translating BraTS Brain Tumor Segmentation Algorithms Into Clinical and Scientific Practice

Despite great advances in brain tumor segmentation and clear clinical need, translation of state-of-the-art computational methods into clinical routine and scientific practice remains a major challenge. Several factors impede successful implementations, including data standardization and preprocessing. However, these steps are pivotal for the deployment of state-of-the-art image segmentation algorithms. To overcome these issues, we present BraTS Toolkit. BraTS Toolkit is a holistic approach to brain tumor segmentation and consists of three components: First, the BraTS Preprocessor facilitates data standardization and preprocessing for researchers and clinicians alike. It covers the entire image analysis workflow prior to tumor segmentation, from image conversion and registration to brain extraction. Second, BraTS Segmentor enables orchestration of BraTS brain tumor segmentation algorithms for generation of fully-automated segmentations. Finally, Brats Fusionator can combine the resulting candidate segmentations into consensus segmentations using fusion methods such as majority voting and iterative SIMPLE fusion. The capabilities of our tools are illustrated with a practical example to enable easy translation to clinical and scientific practice.

Deep convolutional neural networks for automated segmentation of brain metastases trained on clinical data

INTRODUCTION

Deep learning-based algorithms have demonstrated enormous performance in segmentation of medical images. We collected a dataset of multiparametric MRI and contour data acquired for use in radiosurgery, to evaluate the performance of deep convolutional neural networks (DCNN) in automatic segmentation of brain metastases (BM).

METHODS

A conventional U-Net (cU-Net), a modified U-Net (moU-Net) and a U-Net trained only on BM smaller than 0.4 ml (sU-Net) were implemented. Performance was assessed on a separate test set employing sensitivity, specificity, average false positive rate (AFPR), the dice similarity coefficient (DSC), Bland-Altman analysis and the concordance correlation coefficient (CCC).

RESULTS

A dataset of 509 patients (1223 BM) was split into a training set (469 pts) and a test set (40 pts). A combination of all trained networks was the most sensitive (0.82) while maintaining a specificity 0.83. The same model achieved a sensitivity of 0.97 and a specificity of 0.94 when considering only lesions larger than 0.06 ml (75% of all lesions). Type of primary cancer had no significant influence on the mean DSC per lesion (p = 0.60). Agreement between manually and automatically assessed tumor volumes as quantified by a CCC of 0.87 (95% CI, 0.77-0.93), was excellent.

CONCLUSION

Using a dataset which properly captured the variation in imaging appearance observed in clinical practice, we were able to conclude that DCNNs reach clinically relevant performance for most lesions. Clinical applicability is currently limited by the size of the target lesion. Further studies should address if small targets are accurately represented in the test data.

Automated brain extraction of multisequence MRI using artificial neural networks

Brain extraction is a critical preprocessing step in the analysis of neuroimaging studies conducted with magnetic resonance imaging (MRI) and influences the accuracy of downstream analyses. The majority of brain extraction algorithms are, however, optimized for processing healthy brains and thus frequently fail in the presence of pathologically altered brain or when applied to heterogeneous MRI datasets. Here we introduce a new, rigorously validated algorithm (termed HD-BET) relying on artificial neural networks that aim to overcome these limitations. We demonstrate that HD-BET outperforms six popular, publicly available brain extraction algorithms in several large-scale neuroimaging datasets, including one from a prospective multicentric trial in neuro-oncology, yielding state-of-the-art performance with median improvements of +1.16 to +2.50 points for the Dice coefficient and -0.66 to -2.51 mm for the Hausdorff distance. Importantly, the HD-BET algorithm, which shows robust performance in the presence of pathology or treatment-induced tissue alterations, is applicable to a broad range of MRI sequence types and is not influenced by variations in MRI hardware and acquisition parameters encountered in both research and clinical practice. For broader accessibility, the HD-BET prediction algorithm is made freely available (www.neuroAI-HD.org) and may become an essential component for robust, automated, high-throughput processing of MRI neuroimaging data.