Deep learning for automatic segmentation of vestibular schwannoma: a retrospective study from multi-center routine MRI.

Automatic segmentation of vestibular schwannoma (VS) from routine clinical MRI has potential to improve clinical workflow, facilitate treatment decisions, and assist patient management. Previous work demonstrated reliable automatic segmentation performance on datasets of standardized MRI images acquired for stereotactic surgery planning. However, diagnostic clinical datasets are generally more diverse and pose a larger challenge to automatic segmentation algorithms, especially when post-operative images are included. In this work, we show for the first time that automatic segmentation of VS on routine MRI datasets is also possible with high accuracy. We acquired and publicly release a curated multi-center routine clinical (MC-RC) dataset of 160 patients with a single sporadic VS. For each patient up to three longitudinal MRI exams with contrast-enhanced T1-weighted (ceT1w) (n = 124) and T2-weighted (T2w) (n = 363) images were included and the VS manually annotated. Segmentations were produced and verified in an iterative process: (1) initial segmentations by a specialized company; (2) review by one of three trained radiologists; and (3) validation by an expert team. Inter- and intra-observer reliability experiments were performed on a subset of the dataset. A state-of-the-art deep learning framework was used to train segmentation models for VS. Model performance was evaluated on a MC-RC hold-out testing set, another public VS datasets, and a partially public dataset. The generalizability and robustness of the VS deep learning segmentation models increased significantly when trained on the MC-RC dataset. Dice similarity coefficients (DSC) achieved by our model are comparable to those achieved by trained radiologists in the inter-observer experiment. On the MC-RC testing set, median DSCs were 86.2(9.5) for ceT1w, 89.4(7.0) for T2w, and 86.4(8.6) for combined ceT1w+T2w input images. On another public dataset acquired for Gamma Knife stereotactic radiosurgery our model achieved median DSCs of 95.3(2.9), 92.8(3.8), and 95.5(3.3), respectively. In contrast, models trained on the Gamma Knife dataset did not generalize well as illustrated by significant underperformance on the MC-RC routine MRI dataset, highlighting the importance of data variability in the development of robust VS segmentation models. The MC-RC dataset and all trained deep learning models were made available online.

CrossMoDA 2021 challenge: Benchmark of cross-modality domain adaptation techniques for vestibular schwannoma and cochlea segmentation.

Domain Adaptation (DA) has recently been of strong interest in the medical imaging community. While a large variety of DA techniques have been proposed for image segmentation, most of these techniques have been validated either on private datasets or on small publicly available datasets. Moreover, these datasets mostly addressed single-class problems. To tackle these limitations, the Cross-Modality Domain Adaptation (crossMoDA) challenge was organised in conjunction with the 24th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2021). CrossMoDA is the first large and multi-class benchmark for unsupervised cross-modality Domain Adaptation. The goal of the challenge is to segment two key brain structures involved in the follow-up and treatment planning of vestibular schwannoma (VS): the VS and the cochleas. Currently, the diagnosis and surveillance in patients with VS are commonly performed using contrast-enhanced T1 (ceT1) MR imaging. However, there is growing interest in using non-contrast imaging sequences such as high-resolution T2 (hrT2) imaging. For this reason, we established an unsupervised cross-modality segmentation benchmark. The training dataset provides annotated ceT1 scans (N=105) and unpaired non-annotated hrT2 scans (N=105). The aim was to automatically perform unilateral VS and bilateral cochlea segmentation on hrT2 scans as provided in the testing set (N=137). This problem is particularly challenging given the large intensity distribution gap across the modalities and the small volume of the structures. A total of 55 teams from 16 countries submitted predictions to the validation leaderboard. Among them, 16 teams from 9 different countries submitted their algorithm for the evaluation phase. The level of performance reached by the top-performing teams is strikingly high (best median Dice score - VS: 88.4%; Cochleas: 85.7%) and close to full supervision (median Dice score - VS: 92.5%; Cochleas: 87.7%). All top-performing methods made use of an image-to-image translation approach to transform the source-domain images into pseudo-target-domain images. A segmentation network was then trained using these generated images and the manual annotations provided for the source image.

Automated Koos Classification of Vestibular Schwannoma.

Objective: The Koos grading scale is a frequently used classification system for vestibular schwannoma (VS) that accounts for extrameatal tumor dimension and compression of the brain stem. We propose an artificial intelligence (AI) pipeline to fully automate the segmentation and Koos classification of VS from MRI to improve clinical workflow and facilitate patient management. Methods: We propose a method for Koos classification that does not only rely on available images but also on automatically generated segmentations. Artificial neural networks were trained and tested based on manual tumor segmentations and ground truth Koos grades of contrast-enhanced T1-weighted (ceT1) and high-resolution T2-weighted (hrT2) MR images from subjects with a single sporadic VS, acquired on a single scanner and with a standardized protocol. The first stage of the pipeline comprises a convolutional neural network (CNN) which can segment the VS and 7 adjacent structures. For the second stage, we propose two complementary approaches that are combined in an ensemble. The first approach applies a second CNN to the segmentation output to predict the Koos grade, the other approach extracts handcrafted features which are passed to a Random Forest classifier. The pipeline results were compared to those achieved by two neurosurgeons. Results: Eligible patients (n = 308) were pseudo-randomly split into 5 groups to evaluate the model performance with 5-fold cross-validation. The weighted macro-averaged mean absolute error (MA-MAE), weighted macro-averaged F1 score (F1), and accuracy score of the ensemble model were assessed on the testing sets as follows: MA-MAE = 0.11 ± 0.05, F1 = 89.3 ± 3.0%, accuracy = 89.3 ± 2.9%, which was comparable to the average performance of two neurosurgeons: MA-MAE = 0.11 ± 0.08, F1 = 89.1 ± 5.2, accuracy = 88.6 ± 5.8%. Inter-rater reliability was assessed by calculating Fleiss' generalized kappa (k = 0.68) based on all 308 cases, and intra-rater reliabilities of annotator 1 (k = 0.95) and annotator 2 (k = 0.82) were calculated according to the weighted kappa metric with quadratic (Fleiss-Cohen) weights based on 15 randomly selected cases. Conclusions: We developed the first AI framework to automatically classify VS according to the Koos scale. The excellent results show that the accuracy of the framework is comparable to that of neurosurgeons and may therefore facilitate management of patients with VS. The models, code, and ground truth Koos grades for a subset of publicly available images (n = 188) will be released upon publication.

Segmentation of vestibular schwannoma from MRI, an open annotated dataset and baseline algorithm.

Automatic segmentation of vestibular schwannomas (VS) from magnetic resonance imaging (MRI) could significantly improve clinical workflow and assist patient management. We have previously developed a novel artificial intelligence framework based on a 2.5D convolutional neural network achieving excellent results equivalent to those achieved by an independent human annotator. Here, we provide the first publicly-available annotated imaging dataset of VS by releasing the data and annotations used in our prior work. This collection contains a labelled dataset of 484 MR images collected on 242 consecutive patients with a VS undergoing Gamma Knife Stereotactic Radiosurgery at a single institution. Data includes all segmentations and contours used in treatment planning and details of the administered dose. Implementation of our automated segmentation algorithm uses MONAI, a freely-available open-source framework for deep learning in healthcare imaging. These data will facilitate the development and validation of automated segmentation frameworks for VS and may also be used to develop other multi-modal algorithmic models.

Pulse sequences for measuring exchange rates between proton species: From unlocalised NMR spectroscopy to chemical exchange saturation transfer imaging.

Within the field of NMR spectroscopy, the study of chemical exchange processes through saturation transfer techniques has a long history. In the context of MRI, chemical exchange techniques have been adapted to increase the sensitivity of imaging to small fractions of exchangeable protons, including the labile protons of amines, amides and hydroxyls. The MR contrast is generated by frequency-selective irradiation of the labile protons, which results in a reduction of the water signal associated with transfer of the labile protons' saturated magnetization to the protons of the surrounding free water. The signal intensity depends on the rate of chemical exchange and the concentration of labile protons as well as on the properties of the irradiation field. This methodology is referred to as CEST (chemical exchange saturation transfer) imaging. Applications of CEST include imaging of molecules with short transverse relaxation times and mapping of physiological parameters such as pH, temperature, buffer concentration and chemical composition due to the dependency of this chemical exchange effect on all these parameters. This article aims to describe these effects both theoretically and experimentally. In depth analysis and mathematical modelling are provided for all pulse sequences designed to date to measure the chemical exchange rate. Importantly, it has become clear that the background signal from semi-solid protons and the presence of the Nuclear Overhauser Effect (NOE), either through direct dipole-dipole mechanisms or through exchange-relayed signals, complicates the analysis of CEST effects. Therefore, advanced methods to suppress these confounding factors have been developed, and these are also reviewed. Finally, the experimental work conducted both in vitro and in vivo is discussed and the progress of CEST imaging towards clinical practice is presented.

Translating pH-sensitive PROgressive saturation for QUantifying Exchange rates using Saturation Times (PRO-QUEST) MRI to a 3T clinical scanner.

PURPOSE: To translate the recently developed PRO-QUEST (Progressive saturation for quantifying exchange rates using saturation times) sequence from preclinical 9.4T to 3T clinical magnetic field strength. METHODS: Numerical simulations were performed to define the optimal saturation flip angles for PRO-QUEST saturation pulses at 3T and demonstrate the effect of a ∆T2 error on the exchange rate (kex ) estimation at various field strengths. Exchange-dependent relaxation rate (Rex ) was measured for glutamate solutions in various pH, healthy volunteers and patients with multiple sclerosis (MS). Additionally, concentration-independent ratiometric Rex maps were produced to evaluate regional signal variations across the brain of human volunteers. RESULTS: The calculated Rex significantly correlates with pH in glutamate samples, however, kex values are underestimated as compared to those previously obtained at 9.4T. In the ratiometric Rex map of healthy volunteers, no significant differences are found between grey matter, white matter, and basal ganglia. In patients with MS, white matter lesions are visible in single saturation power Rex maps whereas only a periventricular lesion is apparent in the ratiometric Rex map. CONCLUSION: We demonstrate that quantification of pH sensitive indices using PRO-QUEST is feasible at 3T within clinically acceptable acquisition times. Our initial findings in patients with MS show that pH sensitive indices varied with the type of lesion examined whereas no significant difference was found in healthy volunteers between tissue types, suggesting that it would be worthwhile to apply PRO-QUEST in a larger cohort of patients to better understand its distinct imaging features relative to conventional techniques.

Challenges in glucoCEST MR body imaging at 3 Tesla.

BACKGROUND: The aim of this study was to translate dynamic glucose enhancement (DGE) body magnetic resonance imaging (MRI) based on the glucose chemical exchange saturation transfer (glucoCEST) signal to a 3 T clinical field strength. METHODS: An infusion protocol for intravenous (i.v.) glucose was optimised using a hyperglycaemic clamp to maximise the chances of detecting exchange-sensitive MRI signal. Numerical simulations were performed to define the optimum parameters for glucoCEST measurements with consideration to physiological conditions. DGE images were acquired for patients with lymphomas and prostate cancer injected i.v. with 20% glucose. RESULTS: The optimised hyperglycaemic clamp infusion based on the DeFronzo method demonstrated higher efficiency and stability of glucose delivery as compared to manual determination of glucose infusion rates. DGE signal sensitivity was found to be dependent on T2, B1 saturation power and integration range. Our results show that motion correction and B0 field inhomogeneity correction are crucial to avoid mistaking signal changes for a glucose response while field drift is a substantial contributor. However, after B0 field drift correction, no significant glucoCEST signal enhancement was observed in tumour regions of all patients in vivo. CONCLUSIONS: Based on our simulated and experimental results, we conclude that glucose-related signal remains elusive at 3 T in body regions, where physiological movements and strong effects of B1 + and B0 render the originally small glucoCEST signal difficult to detect.

Optimization and repeatability of multipool chemical exchange saturation transfer MRI of the prostate at 3.0 T.

BACKGROUND: Chemical exchange saturation transfer (CEST) can potentially support cancer imaging with metabolically derived information. Multiparametric prostate MRI has improved diagnosis but may benefit from additional information to reduce the need for biopsies. PURPOSE: To optimize an acquisition and postprocessing protocol for 3.0 T multipool CEST analysis of prostate data and evaluate the repeatability of the technique. STUDY TYPE: Prospective. SUBJECTS: Five healthy volunteers (age range: 24-47 years; median age: 28 years) underwent two sessions (interval range: 7-27 days; median interval: 20 days) and two biopsy-proven prostate cancer patients were evaluated once. Patient 1 (71 years) had a Gleason 3 + 4 transition zone (TZ) tumor and patient 2 (55 years) had a Gleason 4 + 3 peripheral zone (PZ) tumor. FIELD STRENGTH: 3.0 T. Sequences run: T2 -weighted turbo-spin-echo (TSE); diffusion-weighted imaging; CEST; WASABI (for B0 determination). ASSESSMENT: Saturation, readout, and fit-model parameters were optimized to maximize in vivo amide and nuclear Overhauser effect (NOE) signals. Repeatability (intrasession and intersession) was evaluated in healthy volunteers. Subsequently, preliminary evaluation of signal differences was made in patients. Regions of interest were drawn by two post-FRCR board-certified readers, both with over 5 years of experience in multiparametric prostate MRI. STATISTICAL TESTS: Repeatability was assessed using Bland-Altman analysis, coefficient of variation (CV), and 95% limits of agreement (LOA). Statistical significance of CEST contrast was calculated using a nonparametric Mann-Whitney U-test. RESULTS: The optimized saturation scheme was found to be 60 sinc-Gaussian pulses with 40 msec pulse duration, at 50% duty-cycle with continuous-wave pulse equivalent B1 power (B1CWPE ) of 0.92 µT. The magnetization transfer (MT) contribution to the fit-model was centered at -1.27 ppm. Intersession coefficients of variation (CVs) of the amide, NOE, and magnetization transfer (MT) and asymmetric magnetization transfer ratio (MTRasym ) signals of 25%, 23%, 18%, and 200%, respectively, were observed. Fit-metric and MTRasym CVs agreed between readers to within 4 and 10 percentage points, respectively. DATA CONCLUSION: Signal differences of 0.03-0.10 (17-43%) detectable depending upon pool, with MT the most repeatable (signal difference of 17-22% detectable). LEVEL OF EVIDENCE: 2 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2019;50:1238-1250.

Assessment of a clinically feasible Bayesian fitting algorithm using a simplified description of Chemical Exchange Saturation Transfer (CEST) imaging.

Fitting a model based on the Bloch-McConnell (BM) equations to Chemical Exchange Saturation Transfer (CEST) spectra allows for the quantification of metabolite concentration and exchange rate as well as simultaneous correction of field inhomogeneity, direct water saturation and magnetization transfer. Employing a Bayesian fitting approach permits the integration of prior information into the analysis to incorporate expected parameter distributions and to prevent over-fitting. However, the analysis can be time consuming if a general numerical solution of the BM equations is applied. In this study, we combined a Bayesian fitting algorithm with approximate analytical solutions of the BM equations to achieve feasible computational times. To evaluate the accuracy and speed of the suggested approach, phantoms including Iodipamide, Taurine and Creatine were tested in addition to simulated data with continuous-wave (CW) and pulsed saturation with Gaussian pulses. A significant reduction of computational time was achieved when fitting CW data (about 50-fold) and pulsed saturation data (more than 100-fold) with the analytical model while the estimated parameters were largely consistent with the parameters from the general numerical solution. The increased speed of the algorithm facilitates the Bayesian analysis of CEST data within clinically feasible processing times. Other analytical models valid for different parameter regimes may be employed to extend the applicability to a wider range of CEST agents.

Partial volume correction for in vivo (23)Na-MRI data of the human brain.

The concentration of sodium is a functional cell parameter and absolute quantification can be interesting for diagnostical purposes. The accuracy of sodium magnetic resonance imaging ((23)Na-MRI) is strongly biased by partial volume effects (PVEs). Hence our purpose was to establish a partial volume correction (PVC) method for (23)Na-MRI. The existing geometric transfer matrix (GTM) correction method was transferred from positron emission tomography (PET) to (23)Na-MRI and tested in a phantom study. Different parameters, as well as accuracy of registration and segmentation were evaluated prior to first in vivo measurements. In vivo sodium data-sets of the human brain were obtained at B0=7T with a nominal spatial resolution of (3mm)(3) using a density adapted radial pulse sequence. A volunteer study with four healthy subjects was performed to measure partial volume (PV) corrected tissue sodium concentration (TSC) which was verified by means of an intrinsic correction control. In the phantom study the PVC algorithm yielded a good correction performance and reduced the discrepancy between the measured sodium concentration value and the expected value in the smallest compartments of the phantom by 11% to a mean PVE induced discrepancy of 5.7% after correction. The corrected in vivo data showed a reduction of PVE bias for the investigated compartments for all volunteers, resulting in a mean reduction of discrepancy between two separate CSF compartments from 36% to 7.6%. The absolute TSC for two separate CSF compartments (sulci, lateral ventricles), gray and white brain matter after correction were 129±8mmol/L, 138±4mmol/L, 48±1mmol/L and 43±3mmol/L, respectively. The applied PVC algorithm reduces the PV-bias in quantitative (23)Na-MRI. Accurate, high-resolution anatomical data is required to enable appropriate PVC. The algorithm and segmentation approach is robust and leads to reproducible results.