Synthesizing 3D Multi-Contrast Brain Tumor MRIs Using Tumor Mask Conditioning.

Data scarcity and data imbalance are two major challenges in training deep learning models on medical images, such as brain tumor MRI data. The recent advancements in generative artificial intelligence have opened new possibilities for synthetically generating MRI data, including brain tumor MRI scans. This approach can be a potential solution to mitigate the data scarcity problem and enhance training data availability. This work focused on adapting the 2D latent diffusion models to generate 3D multi-contrast brain tumor MRI data with a tumor mask as the condition. The framework comprises two components: a 3D autoencoder model for perceptual compression and a conditional 3D Diffusion Probabilistic Model (DPM) for generating high-quality and diverse multi-contrast brain tumor MRI samples, guided by a conditional tumor mask. Unlike existing works that focused on generating either 2D multi-contrast or 3D single-contrast MRI samples, our models generate multi-contrast 3D MRI samples. We also integrated a conditional module within the UNet backbone of the DPM to capture the semantic class-dependent data distribution driven by the provided tumor mask to generate MRI brain tumor samples based on a specific brain tumor mask. We trained our models using two brain tumor datasets: The Cancer Genome Atlas (TCGA) public dataset and an internal dataset from the University of Texas Southwestern Medical Center (UTSW). The models were able to generate high-quality 3D multi-contrast brain tumor MRI samples with the tumor location aligned by the input condition mask. The quality of the generated images was evaluated using the Fréchet Inception Distance (FID) score. This work has the potential to mitigate the scarcity of brain tumor data and improve the performance of deep learning models involving brain tumor MRI data.

A Two-Stage Training Framework Using Multicontrast MRI Radiomics for IDH Mutation Status Prediction in Glioma.

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To develop a radiomics framework for preoperative MRI-based prediction of IDH mutation status, a crucial glioma prognostic indicator. Materials and Methods Radiomics features (shape, first-order statistics, and texture) were extracted from the whole tumor or the combination of nonenhancing, necrosis, and edema regions. Segmentation masks were obtained via the federated tumor segmentation tool or the original data source. Boruta, a wrapper-based feature selection algorithm, identified relevant features. Addressing the imbalance between mutated and wild-type cases, multiple prediction models were trained on balanced data subsets using Random Forest or XGBoost and assembled to build the final classifier. The framework was evaluated using retrospective MRI scans from three public datasets (The Cancer Imaging Archive (TCIA, 227 patients), the University of California San Francisco Preoperative Diffuse Glioma MRI dataset (UCSF, 495 patients), and the Erasmus Glioma Database (EGD, 456 patients)) and internal datasets collected from UT Southwestern Medical Center (UTSW, 356 patients), New York University (NYU, 136 patients), and University of Wisconsin-Madison (UWM, 174 patients). TCIA and UTSW served as separate training sets, while the remaining data constituted the test set (1617 or 1488 testing cases, respectively). Results The best-performing models trained on the TCIA dataset achieved area under the receiver operating characteristic curve (AUC) values of 0.89 for UTSW, 0.86 for NYU, 0.93 for UWM, 0.94 for UCSF, and 0.88 for EGD test sets. The best-performing models trained on the UTSW dataset achieved slightly higher AUCs: 0.92 for TCIA, 0.88 for NYU, 0.96 for UWM, 0.93 for UCSF, and 0.90 for EGD. Conclusion This MRI radiomics-based framework shows promise for accurate preoperative prediction of IDH mutation status in patients with glioma. Published under a CC BY 4.0 license.

Magnetoencephalography Atlas Viewer for Dipole Localization and Viewing.

Magnetoencephalography (MEG) is a noninvasive neuroimaging technique widely recognized for epilepsy and tumor mapping. MEG clinical reporting requires a multidisciplinary team, including expert input regarding each dipole's anatomic localization. Here, we introduce a novel tool, the "Magnetoencephalography Atlas Viewer" (MAV), which streamlines this anatomical analysis. The MAV normalizes the patient's Magnetic Resonance Imaging (MRI) to the Montreal Neurological Institute (MNI) space, reverse-normalizes MNI atlases to the native MRI, identifies MEG dipole files, and matches dipoles' coordinates to their spatial location in atlas files. It offers a user-friendly and interactive graphical user interface (GUI) for displaying individual dipoles, groups, coordinates, anatomical labels, and a tri-planar MRI view of the patient with dipole overlays. It evaluated over 273 dipoles obtained in clinical epilepsy subjects. Consensus-based ground truth was established by three neuroradiologists, with a minimum agreement threshold of two. The concordance between the ground truth and MAV labeling ranged from 79% to 84%, depending on the normalization method. Higher concordance rates were observed in subjects with minimal or no structural abnormalities on the MRI, ranging from 80% to 90%. The MAV provides a straightforward MEG dipole anatomic localization method, allowing a nonspecialist to prepopulate a report, thereby facilitating and reducing the time of clinical reporting.

Assessing the Performance of Artificial Intelligence Models: Insights from the American Society of Functional Neuroradiology Artificial Intelligence Competition.

BACKGROUND AND PURPOSE: Artificial intelligence (AI) models in radiology are frequently developed and validated using datasets from a single institution and are rarely tested on independent, external datasets, raising questions about their generalizability and applicability in clinical practice. The American Society of Functional Neuroradiology (ASFNR) organized a multi-center AI competition to evaluate the proficiency of developed models in identifying various pathologies on NCCT, assessing age-based normality and estimating medical urgency. MATERIALS AND METHODS: In total, 1201 anonymized, full-head NCCT clinical scans from five institutions were pooled to form the dataset. The dataset encompassed normal studies as well as pathologies including acute ischemic stroke, intracranial hemorrhage, traumatic brain injury, and mass effect (detection of these-task 1). NCCTs were also assessed to determine if findings were consistent with expected brain changes for the patient's age (task 2: age-based normality assessment) and to identify any abnormalities requiring immediate medical attention (task 3: evaluation of findings for urgent intervention). Five neuroradiologists labeled each NCCT, with consensus interpretations serving as the ground truth. The competition was announced online, inviting academic institutions and companies. Independent central analysis assessed each model's performance. Accuracy, sensitivity, specificity, positive and negative predictive values, and receiver operating characteristic (ROC) curves were generated for each AI model, along with the area under the ROC curve (AUROC). RESULTS: 1177 studies were processed by four teams. The median age of patients was 62, with an interquartile range of 33. 19 teams from various academic institutions registered for the competition. Of these, four teams submitted their final results. No commercial entities participated in the competition. For task 1, AUROCs ranged from 0.49 to 0.59. For task 2, two teams completed the task with AUROC values of 0.57 and 0.52. For task 3, teams had little to no agreement with the ground truth. CONCLUSIONS: To assess the performance of AI models in real-world clinical scenarios, we analyzed their performance in the ASFNR AI Competition. The first ASFNR Competition underscored the gap between expectation and reality; the models largely fell short in their assessments. As the integration of AI tools into clinical workflows increases, neuroradiologists must carefully recognize the capabilities, constraints, and consistency of these technologies. Before institutions adopt these algorithms, thorough validation is essential to ensure acceptable levels of performance in clinical settings.ABBREVIATIONS: AI = artificial intelligence; ASFNR = American Society of Functional Neuroradiology; AUROC = area under the receiver operating characteristic curve; DICOM = Digital Imaging and Communications in Medicine; GEE = generalized estimation equation; IQR = interquartile range; NPV = negative predictive value; PPV = positive predictive value; ROC = receiver operating characteristic; TBI = traumatic brain injury.

Synthesizing Contrast-Enhanced MR Images from Noncontrast MR Images Using Deep Learning.

BACKGROUND AND PURPOSE: Recent developments in deep learning methods offer a potential solution to the need for alternative imaging methods due to concerns about the toxicity of gadolinium-based contrast agents. The purpose of the study was to synthesize virtual gadolinium contrast-enhanced T1-weighted MR images from noncontrast multiparametric MR images in patients with primary brain tumors by using deep learning. MATERIALS AND METHODS: We trained and validated a deep learning network by using MR images from 335 subjects in the Brain Tumor Segmentation Challenge 2019 training data set. A held out set of 125 subjects from the Brain Tumor Segmentation Challenge 2019 validation data set was used to test the generalization of the model. A residual inception DenseNet network, called T1c-ET, was developed and trained to simultaneously synthesize virtual contrast-enhanced T1-weighted (vT1c) images and segment the enhancing portions of the tumor. Three expert neuroradiologists independently scored the synthesized vT1c images by using a 3-point Likert scale, evaluating image quality and contrast enhancement against ground truth T1c images (1 = poor, 2 = good, 3 = excellent). RESULTS: The synthesized vT1c images achieved structural similarity index, peak signal-to-noise ratio, and normalized mean square error scores of 0.91, 64.35, and 0.03, respectively. There was moderate interobserver agreement between the 3 raters, regarding the algorithm's performance in predicting contrast enhancement, with a Fleiss kappa value of 0.61. Our model was able to accurately predict contrast enhancement in 88.8% of the cases (scores of 2 to 3 on the 3-point scale). CONCLUSIONS: We developed a novel deep learning architecture to synthesize virtual postcontrast enhancement by using only conventional noncontrast brain MR images. Our results demonstrate the potential of deep learning methods to reduce the need for gadolinium contrast in the evaluation of primary brain tumors.

Magnetoencephalography (MEG) Data Processing in Epilepsy Patients with Implanted Responsive Neurostimulation (RNS) Devices.

Drug-resistant epilepsy (DRE) is often treated with surgery or neuromodulation. Specifically, responsive neurostimulation (RNS) is a widely used therapy that is programmed to detect abnormal brain activity and intervene with tailored stimulation. Despite the success of RNS, some patients require further interventions. However, having an RNS device in situ is a hindrance to the performance of neuroimaging techniques. Magnetoencephalography (MEG), a non-invasive neurophysiologic and functional imaging technique, aids epilepsy assessment and surgery planning. MEG performed post-RNS is complicated by signal distortions. This study proposes an independent component analysis (ICA)-based approach to enhance MEG signal quality, facilitating improved assessment for epilepsy patients with implanted RNS devices. Three epilepsy patients, two with RNS implants and one without, underwent MEG scans. Preprocessing included temporal signal space separation (tSSS) and an automated ICA-based approach with MNE-Python. Power spectral density (PSD) and signal-to-noise ratio (SNR) were analyzed, and MEG dipole analysis was conducted using single equivalent current dipole (SECD) modeling. The ICA-based noise removal preprocessing method substantially improved the signal-to-noise ratio (SNR) for MEG data from epilepsy patients with implanted RNS devices. Qualitative assessment confirmed enhanced signal readability and improved MEG dipole analysis. ICA-based processing markedly enhanced MEG data quality in RNS patients, emphasizing its clinical relevance.

MRI-Based Deep Learning Method for Classification of IDH Mutation Status.

Isocitrate dehydrogenase (IDH) mutation status has emerged as an important prognostic marker in gliomas. This study sought to develop deep learning networks for non-invasive IDH classification using T2w MR images while comparing their performance to a multi-contrast network. Methods: Multi-contrast brain tumor MRI and genomic data were obtained from The Cancer Imaging Archive (TCIA) and The Erasmus Glioma Database (EGD). Two separate 2D networks were developed using nnU-Net, a T2w-image-only network (T2-net) and a multi-contrast network (MC-net). Each network was separately trained using TCIA (227 subjects) or TCIA + EGD data (683 subjects combined). The networks were trained to classify IDH mutation status and implement single-label tumor segmentation simultaneously. The trained networks were tested on over 1100 held-out datasets including 360 cases from UT Southwestern Medical Center, 136 cases from New York University, 175 cases from the University of Wisconsin-Madison, 456 cases from EGD (for the TCIA-trained network), and 495 cases from the University of California, San Francisco public database. A receiver operating characteristic curve (ROC) was drawn to calculate the AUC value to determine classifier performance. Results: T2-net trained on TCIA and TCIA + EGD datasets achieved an overall accuracy of 85.4% and 87.6% with AUCs of 0.86 and 0.89, respectively. MC-net trained on TCIA and TCIA + EGD datasets achieved an overall accuracy of 91.0% and 92.8% with AUCs of 0.94 and 0.96, respectively. We developed reliable, high-performing deep learning algorithms for IDH classification using both a T2-image-only and a multi-contrast approach. The networks were tested on more than 1100 subjects from diverse databases, making this the largest study on image-based IDH classification to date.

ELectronic Archimedes spiral Neural Network (ELANN).

The Archimedes spiral is a clinical tool that aids in the diagnosis and monitoring of essential tremor. However, spiral ratings may vary based on experience and training of the rating physician. This study sought to generate an objective standard model for tremor evaluation using convolutional neural networks. One senior movement disorders neurologist (Neurologist 1) with over 30 years of clinical experience used the Bain and Findley Spirography Rating Scale to rate 1653 Archimedes spiral images from 46 essential tremor patients (mild to severe tremor) and 75 control subjects (no to mild tremor). Neurologist 1's labels were used as the reference standard to train the model. After training the model, a randomly selected subset of spiral testing data was re-evaluated by Neurologist 1, by a second senior movement disorders neurologist (Neurologist 2) with over 27 years of clinical experience, and by our model. Cohen's Weighted Kappa 95% confidence intervals were calculated from all rater comparisons to determine if our model performs with the same proficiency as two senior movement disorders neurologists. The Cohen's Weighted Kappa 95% confidence intervals for the agreement between the reference standard scores and Neurologist 1's rerated scores, for the agreement between the reference standard scores and Neurologist 2's scores, and for the agreement between the reference standard scores and our model's scores were 0.93-0.98, 0.86-0.94, and 0.89-0.96, respectively. With overlapping Cohen's Weighted Kappa 95% confidence intervals for all agreement comparisons, we demonstrate that our model evaluates spirals with the same proficiency as two senior movement disorders neurologists.

Improving tremor response to focused ultrasound thalamotomy.

MRI-guided high-intensity focused ultrasound thalamotomy is an incisionless therapy for essential tremor. To reduce adverse effects, the field has migrated to treating at 2 mm above the anterior commissure-posterior commissure plane. We perform MRI-guided high-intensity focused ultrasound with an advanced imaging targeting technique, four-tract tractography. Four-tract tractography uses diffusion tensor imaging to identify the critical white matter targets for tremor control, the decussating and non-decussating dentatorubrothalamic tracts, while the corticospinal tract and medial lemniscus are identified to be avoided. In some patients, four-tract tractography identified a risk of damaging the medial lemniscus or corticospinal tract if treated at 2 mm superior to the anterior commissure-posterior commissure plane. In these patients, we chose to target 1.2-1.5 mm superior to the anterior commissure-posterior commissure plane. In these patients, post-operative imaging revealed that the focused ultrasound lesion extended into the posterior subthalamic area. This study sought to determine if patients with focused ultrasound lesions that extend into the posterior subthalamic area have a differnce in tremor improvement than those without. Twenty essential tremor patients underwent MRI-guided high-intensity focused ultrasound and were retrospectively classified into two groups. Group 1 included patients with an extension of the thalamic-focused ultrasound lesion into the posterior subthalamic area. Group 2 included patients without extension of the thalamic-focused ultrasound lesion into the posterior subthalamic area. For each patient, the percent change in postural tremor, kinetic tremor and Archimedes spiral scores were calculated between baseline and a 3-month follow-up. Two-tailed Wilcoxon rank-sum tests were used to compare the improvement in tremor scores, the total number of sonications, thermal dose to achieve initial tremor response, and skull density ratio between groups. Group 1 had significantly greater postural, kinetic, and Archimedes spiral score percent improvement than Group 2 (P values: 5.41 × 10-5, 4.87 × 10-4, and 5.41 × 10-5, respectively). Group 1 also required significantly fewer total sonications to control the tremor and a significantly lower thermal dose to achieve tremor response (P values: 6.60 × 10-4 and 1.08 × 10-5, respectively). No significant group differences in skull density ratio were observed (P = 1.0). We do not advocate directly targeting the posterior subthalamic area with MRI-guided high-intensity focused ultrasound because the shape of the focused ultrasound lesion can result in a high risk of adverse effects. However, when focused ultrasound lesions naturally extend from the thalamus into the posterior subthalamic area, they provide greater tremor control than those that only involve the thalamus.

Establishing language laterality: does resting-state functional MRI help?

OBJECTIVE: Task-based functional MRI (tb-fMRI) is now considered the standard, noninvasive technique in establishing language laterality in children for surgical planning. The evaluation can be limited due to several factors such as age, language barriers, and developmental and cognitive delays. Resting-state functional MRI (rs-fMRI) offers a potential path to establish language dominance without active task participation. The authors sought to compare the ability of rs-fMRI for language lateralization in the pediatric population with conventional tb-fMRI used as the gold standard. METHODS: The authors performed a retrospective evaluation of all pediatric patients at a dedicated quaternary pediatric hospital who underwent tb-fMRI and rs-fMRI from 2019 to 2021 as part of the surgical workup for patients with seizures and brain tumors. Task-based fMRI language laterality was based on a patient's adequate performance on one or more of the following: sentence completion, verb generation, antonym generation, or passive listening tasks. Resting-state fMRI data were postprocessed using statistical parametric mapping, FMRIB Software Library, and FreeSurfer as described in the literature. The laterality index (LI) was calculated from the independent component (IC) with the highest Jaccard Index (JI) for the language mask. Additionally, the authors visually inspected the activation maps for two ICs with the highest JIs. The rs-fMRI LI of IC1 and the authors' image-based subjective interpretation of language lateralization were compared with tb-fMRI, which was considered the gold standard for this study. RESULTS: A retrospective search yielded 33 patients with language fMRI data. Eight patients were excluded (5 with suboptimal tb-fMRI and 3 with suboptimal rs-fMRI data). Twenty-five patients (age range 7-19 years, male/female ratio 15:10) were included in the study. The language laterality concordance between tb-fMRI and rs-fMRI ranged from 68% to 80% for assessment based on LI of independent component analysis with highest JI and for subjective evaluation by visual inspection of activation maps, respectively. CONCLUSIONS: The concordance rates between tb-fMRI and rs-fMRI of 68% to 80% show the limitation of rs-fMRI in determining language dominance. Resting-state fMRI should not be used as the sole method for language lateralization in clinical practice.

Errata: Brain tumor IDH, 1p/19q, and MGMT molecular classification using MRI-based deep learning: an initial study on the effect of motion and motion correction.

[This corrects the article DOI: 10.1117/1.JMI.9.1.016001.].

Corrigendum to: A novel fully automated MRI-based deep-learning method for classification of 1p/19q co-deletion status in brain gliomas.

[This corrects the article DOI: 10.1093/noajnl/vdaa066.].

MEG measured delta waves increase in adolescents after concussion.

INTRODUCTION: The purpose of this study is to determine if delta waves, measured by magnetoencephalography (MEG), increase in adolescents due to a sports concussion. METHODS: Twenty-four adolescents (age 14-17) completed pre- and postseason MRI and MEG scanning. MEG whole-brain delta power was calculated for each subject and normalized by the subject's total power. In eight high school football players diagnosed with a concussion during the season (mean age = 15.8), preseason delta power was subtracted from their postseason scan. In eight high school football players without a concussion (mean age = 15.7), preseason delta power was subtracted from postseason delta power and in eight age-matched noncontact controls (mean age = 15.9), baseline delta power was subtracted from a 4-month follow-up scan. ANOVA was used to compare the mean differences between preseason and postseason scans for the three groups of players, with pairwise comparisons based on Student's t-test method. RESULTS: Players with concussions had significantly increased delta wave power at their postseason scans than nonconcussed players (p = .018) and controls (p = .027). CONCLUSION: We demonstrate that a single concussion during the season in adolescent subjects can increase MEG measured delta frequency power at their postseason scan. This adds to the growing body of literature indicating increased delta power following a concussion.

Relationship between Aggressiveness, Self-Confidence, and Perceived Coach Support and Head Impact Exposure in Youth Football.

This study evaluated head impact exposure (HIE) metrics in relation to individual-level determinants of HIE. Youth (n = 13) and high school (n = 21) football players were instrumented with the Head Impact Telemetry (HIT) system during one season. Players completed the Trait-Robustness of Self-Confidence Inventory (TROSCI), Sports Climate Questionnaire (SCQ), and Competitive Aggressiveness and Anger Scale (CAAS), measuring self-confidence, perceived coach support, and competitive aggressiveness, respectively. Relationships between HIE metrics (number of impacts, median and 95th percentile accelerations, and risk-weighted exposure (RWE)) and survey scores were evaluated using linear regression analysis. For middle school athletes, TROSCI scores were significantly negatively associated with the number of competition impacts and the mean number of impacts per player per competition. SCQ scores were significantly positively associated with median linear acceleration during practice. CAAS scores were not significantly associated with biomechanical metrics at either level of play. Perceived coach support and self-confidence might influence HIE among middle school football players. Football athletes' competitive aggressiveness may have less influence their HIE than other factors.

Brain tumor IDH, 1p/19q, and MGMT molecular classification using MRI-based deep learning: an initial study on the effect of motion and motion correction.

Purpose: Deep learning has shown promise for predicting the molecular profiles of gliomas using MR images. Prior to clinical implementation, ensuring robustness to real-world problems, such as patient motion, is crucial. The purpose of this study is to perform a preliminary evaluation on the effects of simulated motion artifact on glioma marker classifier performance and determine if motion correction can restore classification accuracies. Approach: T2w images and molecular information were retrieved from the TCIA and TCGA databases. Simulated motion was added in the k-space domain along the phase encoding direction. Classifier performance for IDH mutation, 1p/19q co-deletion, and MGMT methylation was assessed over the range of 0% to 100% corrupted k-space lines. Rudimentary motion correction networks were trained on the motion-corrupted images. The performance of the three glioma marker classifiers was then evaluated on the motion-corrected images. Results: Glioma marker classifier performance decreased markedly with increasing motion corruption. Applying motion correction effectively restored classification accuracy for even the most motion-corrupted images. For isocitrate dehydrogenase (IDH) classification, 99% accuracy was achieved, exceeding the original performance of the network and representing a new benchmark in non-invasive MRI-based IDH classification. Conclusions: Robust motion correction can facilitate highly accurate deep learning MRI-based molecular marker classification, rivaling invasive tissue-based characterization methods. Motion correction may be able to increase classification accuracy even in the absence of a visible artifact, representing a new strategy for boosting classifier performance.

Cumulative strain-based metrics for predicting subconcussive head impact exposure-related imaging changes in a cohort of American youth football players.

OBJECTIVE: Youth football athletes are exposed to repetitive subconcussive head impacts during normal participation in the sport, and there is increasing concern about the long-term effects of these impacts. The objective of the current study was to determine if strain-based cumulative exposure measures are superior to kinematic-based exposure measures for predicting imaging changes in the brain. METHODS: This prospective, longitudinal cohort study was conducted from 2012 to 2017 and assessed youth, male football athletes. Kinematic data were collected at all practices and games from enrolled athletes participating in local youth football organizations in Winston-Salem, North Carolina, and were used to calculate multiple risk-weighted cumulative exposure (RWE) kinematic metrics and 36 strain-based exposure metrics. Pre- and postseason imaging was performed at Wake Forest School of Medicine, and diffusion tensor imaging (DTI) measures, including fractional anisotropy (FA), and its components (CL, CP, and CS), and mean diffusivity (MD), were investigated. Included participants were youth football players ranging in age from 9 to 13 years. Exclusion criteria included any history of previous neurological illness, psychiatric illness, brain tumor, concussion within the past 6 months, and/or contraindication to MRI. RESULTS: A total of 95 male athletes (mean age 11.9 years [SD 1.0 years]) participated between 2012 and 2017, with some participating for multiple seasons, resulting in 116 unique athlete-seasons. Regression analysis revealed statistically significant linear relationships between the FA, linear coefficient (CL), and spherical coefficient (CS) and all strain exposure measures, and well as the planar coefficient (CP) and 8 strain measures. For the kinematic exposure measures, there were statistically significant relationships between FA and RWE linear (RWEL) and RWE combined probability (RWECP) as well as CS and RWEL. According to area under the receiver operating characteristic (ROC) curve (AUC) analysis, the best-performing metrics were all strain measures, and included metrics based on tensile, compressive, and shear strain. CONCLUSIONS: Using ROC curves and AUC analysis, all exposure metrics were ranked in order of performance, and the results demonstrated that all the strain-based metrics performed better than any of the kinematic metrics, indicating that strain-based metrics are better discriminators of imaging changes than kinematic-based measures. Studies relating the biomechanics of head impacts with brain imaging and cognitive function may allow equipment designers, care providers, and organizations to prevent, identify, and treat injuries in order to make football a safer activity.

QU-BraTS: MICCAI BraTS 2020 Challenge on Quantifying Uncertainty in Brain Tumor Segmentation - Analysis of Ranking Scores and Benchmarking Results.

Deep learning (DL) models have provided state-of-the-art performance in various medical imaging benchmarking challenges, including the Brain Tumor Segmentation (BraTS) challenges. However, the task of focal pathology multi-compartment segmentation (e.g., tumor and lesion sub-regions) is particularly challenging, and potential errors hinder translating DL models into clinical workflows. Quantifying the reliability of DL model predictions in the form of uncertainties could enable clinical review of the most uncertain regions, thereby building trust and paving the way toward clinical translation. Several uncertainty estimation methods have recently been introduced for DL medical image segmentation tasks. Developing scores to evaluate and compare the performance of uncertainty measures will assist the end-user in making more informed decisions. In this study, we explore and evaluate a score developed during the BraTS 2019 and BraTS 2020 task on uncertainty quantification (QU-BraTS) and designed to assess and rank uncertainty estimates for brain tumor multi-compartment segmentation. This score (1) rewards uncertainty estimates that produce high confidence in correct assertions and those that assign low confidence levels at incorrect assertions, and (2) penalizes uncertainty measures that lead to a higher percentage of under-confident correct assertions. We further benchmark the segmentation uncertainties generated by 14 independent participating teams of QU-BraTS 2020, all of which also participated in the main BraTS segmentation task. Overall, our findings confirm the importance and complementary value that uncertainty estimates provide to segmentation algorithms, highlighting the need for uncertainty quantification in medical image analyses. Finally, in favor of transparency and reproducibility, our evaluation code is made publicly available at https://github.com/RagMeh11/QU-BraTS.

Relationship Between Time-Weighted Head Impact Exposure on Directional Changes in Diffusion Imaging in Youth Football Players.

Approximately 3.5 million youth and adolescents in the US play football, a sport with one of the highest rates of concussion. Repeated subconcussive head impact exposure (HIE) may lead to negative neurological sequelae. To understand HIE as an independent predictive variable, quantitative cumulative kinematic metrics have been developed to capture the volume (i.e., number), severity (i.e., magnitude), and frequency (i.e., time-weighting by the interval between head impacts). In this study, time-weighted cumulative HIE metrics were compared with directional changes in diffusion tensor imaging (DTI) metrics. Changes in DTI conducted on a per-season, per-player basis were assessed as a dependent variable. Directional changes were defined separately as increases and decreases in the number of abnormal voxels relative to non-contact sport controls. Biomechanical and imaging data from 117 athletes (average age 11.9 ± 1.0 years) enrolled in this study was analyzed. Cumulative HIE metrics were more strongly correlated with increases in abnormal voxels than decreases in abnormal voxels. Additionally, across DTI sub-measures, increases and decreases in mean diffusivity (MD) had the strongest relationships with HIE metrics (increases in MD: average R2 = 0.1753, average p = 0.0002; decreases in MD: average R2 = 0.0997, average p = 0.0073). This encourages further investigation into the physiological phenomena represented by directional changes.

MEGnet: Automatic ICA-based artifact removal for MEG using spatiotemporal convolutional neural networks.

Magnetoencephalography (MEG) is a functional neuroimaging tool that records the magnetic fields induced by neuronal activity; however, signal from non-neuronal sources can corrupt the data. Eye-blinks, saccades, and cardiac activity are three of the most common sources of non-neuronal artifacts. They can be measured by affixing eye proximal electrodes, as in electrooculography (EOG), and chest electrodes, as in electrocardiography (ECG), however this complicates imaging setup, decreases patient comfort, and can induce further artifacts from movement. This work proposes an EOG- and ECG-free approach to identify eye-blinks, saccades, and cardiac activity signals for automated artifact suppression. The contribution of this work is three-fold. First, using a data driven, multivariate decomposition approach based on Independent Component Analysis (ICA), a highly accurate artifact classifier is constructed as an amalgam of deep 1-D and 2-D Convolutional Neural Networks (CNNs) to automate the identification and removal of ubiquitous whole brain artifacts including eye-blink, saccade, and cardiac artifacts. The specific architecture of this network is optimized through an unbiased, computer-based hyperparameter random search. Second, visualization methods are applied to the learned abstraction to reveal what features the model uses and to bolster user confidence in the model's training and potential for generalization. Finally, the model is trained and tested on both resting-state and task MEG data from 217 subjects, and achieves a new state-of-the-art in artifact detection accuracy of 98.95% including 96.74% sensitivity and 99.34% specificity on the held out test-set. This work automates MEG processing for both clinical and research use, adapts to the acquired acquisition time, and can obviate the need for EOG or ECG electrodes for artifact detection.

Analysis of longitudinal head impact exposure and white matter integrity in returning youth football players.

OBJECTIVE: The objective of this study was to characterize changes in head impact exposure (HIE) across multiple football seasons and to determine whether changes in HIE correlate with changes in imaging metrics in youth football players. METHODS: On-field head impact data and pre- and postseason imaging data, including those produced by diffusion tensor imaging (DTI), were collected from youth football athletes with at least two consecutive seasons of data. ANCOVA was used to evaluate HIE variations (number of impacts, peak linear and rotational accelerations, and risk-weighted cumulative exposure) by season number. DTI scalar metrics, including fractional anisotropy, mean diffusivity, and linear, planar, and spherical anisotropy coefficients, were evaluated. A control group was used to determine the number of abnormal white matter voxels, which were defined as 2 standard deviations above or below the control group mean. The difference in the number of abnormal voxels between consecutive seasons was computed for each scalar metric and athlete. Linear regression analyses were performed to evaluate relationships between changes in HIE metrics and changes in DTI scalar metrics. RESULTS: There were 47 athletes with multiple consecutive seasons of HIE, and corresponding imaging data were available in a subsample (n = 19) of these. Increases and decreases in HIE metrics were observed among individual athletes from one season to the next, and no significant differences (all p > 0.05) in HIE metrics were observed by season number. Changes in the number of practice impacts, 50th percentile impacts per practice session, and 50th percentile impacts per session were significantly positively correlated with changes in abnormal voxels for all DTI metrics. CONCLUSIONS: These results demonstrate a significant positive association between changes in HIE metrics and changes in the numbers of abnormal voxels between consecutive seasons of youth football. Reducing the number and frequency of head impacts, especially during practice sessions, may decrease the number of abnormal imaging findings from one season to the next in youth football.