Glioblastoma Segmentation Research Articles

Deep Learning-Based Postoperative Glioblastoma Segmentation and Extent of Resection Evaluation: Development, External Validation, and Model Comparison

Abstract Background The pursuit of automated methods to assess the extent of resection (EOR) in glioblastomas is challenging, requiring precise measurement of residual tumor volume. Many algorithms focus on preoperative scans, making them unsuitable for postoperative studies. Our objective was to develop a deep learning-based model for postoperative segmentation using magnetic resonance imaging (MRI). We also compared our model's performance with other available algorithms. Methods To develop the segmentation model, a training cohort from three research institutions and three public databases was used. Multiparametric MRI scans with ground truth labels for contrast enhancing tumor, edema, and surgical cavity, served as training data. The models were trained using MONAI and nnU-Net frameworks. Comparisons were made with currently available segmentation models using an external cohort from a research institution and a public database. Additionally, the model's ability to classify EOR was evaluated using the RANO-Resect classification system. To further validate our best-trained model, an additional independent cohort was used. Results The study included 586 scans: 395 for model training, 52 for model comparison, and 139 scans for independent validation. The nnU-Net framework produced the best model with median Dice scores of 0.81 for contrast enhancing tumor, 0.77 for edema, and 0.81 for surgical cavity. Our best-trained model classified patients into maximal and submaximal resection categories with 96% accuracy in the model comparison dataset and 84% in the independent validation cohort. Conclusion Our nnU-Net-based model outperformed other algorithms in both segmentation and EOR classification tasks, providing a freely accessible tool with promising clinical applicability.

A Comprehensive Review of the Recent Advancements in Imaging Segmentation and Registration Techniques for Glioblastoma and Focusing on the Utilization of Magnetic Resonance Imaging (MRI) and Computed Tomography (CT) Scans.

The most common primary malignant brain tumor is glioblastoma. Glioblastoma Multiforme (GBM) diagnosis is difficult. However, image segmentation and registration methods may simplify and automate Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) scan analysis. Medical practitioners and researchers can better identify and characterize glioblastoma tumors using this technology. Many segmentation and registration approaches have been proposed recently. Note that these approaches are not fully compiled. This review efficiently and critically evaluates the state-of-the-art segmentation and registration techniques for MRI and CT GBM images, providing researchers, medical professionals, and students with a wealth of knowledge to advance GBM imaging and inform decision-making. GBM's origins and development have been examined, along with medical imaging methods used to diagnose tumors. Image segmentation and registration were examined, showing their importance in this difficult task. Frequently encountered glioblastoma segmentation and registration issues were examined. Based on these theoretical foundations, recent image segmentation and registration advances were critically analyzed. Additionally, evaluation measures for analytical efforts were thoroughly reviewed.

Atten‐SEVNETR for volumetric segmentation of glioblastoma and interactive refinement to limit over‐segmentation

AbstractPrecise localization and volumetric segmentation of glioblastoma before and after surgery are crucial for various clinical purposes, including post‐surgery treatment planning, monitoring tumour recurrence, and creating radiotherapy maps. Manual delineation is time‐consuming and prone to errors, hence the adoption of automated 3D quantification methods using deep learning algorithms from MRI scans in recent times. However, automated segmentation often leads to over‐segmentation or under‐segmentation of tumour regions. Introducing an interactive deep‐learning tool would empower radiologists to rectify these inaccuracies by adjusting the over‐segmented and under‐segmented voxels as needed. This paper proposes a network named Atten‐SEVNETR, that has a combined architecture of vision transformers and convolutional neural networks (CNN). This hybrid architecture helps to learn the input volume representation in sequences and focuses on the global multi‐scale information. An interactive graphical user interface is also developed where the initial 3D segmentation of glioblastoma can be interactively corrected to remove falsely detected spurious tumour regions. Atten‐SEVNETR is trained on BraTS training dataset and tested on BraTS validation dataset and on Uppsala University post‐operative glioblastoma dataset. The methodology outperformed state‐of‐the‐art networks like nnFormer, SwinUNet, and SwinUNETR. The mean dice score achieved is 0.7302, and the mean Hausdorff distance‐95 got is 7.78 mm for the Uppsala University dataset.

Exploring the Trade-Off between generalist and specialized Models: A center-based comparative analysis for glioblastoma segmentation

IntroductionInherent variations between inter-center data can undermine the robustness of segmentation models when applied at a specific center (dataset shift). We investigated whether specialized center-specific models are more effective compared to generalist models based on multi-center data, and how center-specific data could enhance the performance of generalist models within a particular center using a fine-tuning transfer learning approach. For this purpose, we studied the dataset shift at center level and conducted a comparative analysis to assess the impact of data source on glioblastoma segmentation models. Methods & MaterialsThe three key components of dataset shift were studied: prior probability shift—variations in tumor size or tissue distribution among centers; covariate shift—inter-center MRI alterations; and concept shift—different criteria for tumor segmentation. BraTS 2021 dataset was used, which includes 1251 cases from 23 centers. Thereafter, 155 deep-learning models were developed and compared, including 1) generalist models trained with multi-center data, 2) specialized models using only center-specific data, and 3) fine-tuned generalist models using center-specific data. ResultsThe three key components of dataset shift were characterized. The amount of covariate shift was substantial, indicating large variations in MR imaging between different centers. Glioblastoma segmentation models tend to perform best when using data from the application center. Generalist models, trained with over 700 samples, achieved a median Dice score of 88.98%. Specialized models surpassed this with 200 cases, while fine-tuned models outperformed with 50 cases. ConclusionsThe influence of dataset shift on model performance is evident. Fine-tuned and specialized models, utilizing data from the evaluated center, outperform generalist models, which rely on data from other centers. These approaches could encourage medical centers to develop customized models for their local use, enhancing the accuracy and reliability of glioblastoma segmentation in a context where dataset shift is inevitable.

Modality redundancy for MRI-based glioblastoma segmentation

PurposeAutomated glioblastoma segmentation from magnetic resonance imaging is generally performed on a four-modality input, including T1, contrast T1, T2 and FLAIR. We hypothesize that information redundancy is present within these image combinations, which can possibly reduce a model’s performance. Moreover, for clinical applications, the risk of encountering missing data rises as the number of required input modalities increases. Therefore, this study aimed to explore the relevance and influence of the different modalities used for MRI-based glioblastoma segmentation.MethodsAfter the training of multiple segmentation models based on nnU-Net and SwinUNETR architectures, differing only in their amount and combinations of input modalities, each model was evaluated with regard to segmentation accuracy and epistemic uncertainty.ResultsResults show that T1CE-based segmentation (for enhanced tumor and tumor core) and T1CE-FLAIR-based segmentation (for whole tumor and overall segmentation) can reach segmentation accuracies comparable to the full-input version. Notably, the highest segmentation accuracy for nnU-Net was found for a three-input configuration of T1CE-FLAIR-T1, suggesting the confounding effect of redundant input modalities. The SwinUNETR architecture appears to suffer less from this, where said three-input and the full-input model yielded statistically equal results.ConclusionThe T1CE-FLAIR-based model can therefore be considered as a minimal-input alternative to the full-input configuration. Addition of modalities beyond this does not statistically improve and can even deteriorate accuracy, but does lower the segmentation uncertainty.

Marker-controlled watershed algorithm and fuzzy C-means clustering machine learning: automated segmentation of glioblastoma from MRI images in a case series.

Automated segmentation of glioblastoma multiforme (GBM) from MRI images is crucial for accurate diagnosis and treatment planning. This paper presents a new and innovative approach for automating the segmentation of GBM from MRI images using the marker-controlled watershed segmentation (MCWS) algorithm. The technique involves several image processing techniques, including adaptive thresholding, morphological filtering, gradient magnitude calculation, and regional maxima identification. The MCWS algorithm efficiently segments images based on local intensity structures using the watershed transform, and fuzzy c-means (FCM) clustering improves segmentation accuracy. The presented approach achieved improved segmentation accuracy in detecting and segmenting GBM tumours from axial T2-weighted (T2-w) MRI images, as demonstrated by the mean characteristics performance metrics for GBM segmentation (sensitivity: 0.9905, specificity: 0.9483, accuracy: 0.9508, precision: 0.5481, F_measure: 0.7052, and jaccard: 0.9340). The results of this study underline the importance of reliable and accurate image segmentation for effective diagnosis and treatment planning of GBM tumours. The MCWS technique provides an effective and efficient approach for the segmentation of challenging medical images.

Deep learning-based algorithm for postoperative glioblastoma MRI segmentation: a promising new tool for tumor burden assessment

ObjectiveClinical and surgical decisions for glioblastoma patients depend on a tumor imaging-based evaluation. Artificial Intelligence (AI) can be applied to magnetic resonance imaging (MRI) assessment to support clinical practice, surgery planning and prognostic predictions. In a real-world context, the current obstacles for AI are low-quality imaging and postoperative reliability. The aim of this study is to train an automatic algorithm for glioblastoma segmentation on a clinical MRI dataset and to obtain reliable results both pre- and post-operatively.MethodsThe dataset used for this study comprises 237 (71 preoperative and 166 postoperative) MRIs from 71 patients affected by a histologically confirmed Grade IV Glioma. The implemented U-Net architecture was trained by transfer learning to perform the segmentation task on postoperative MRIs. The training was carried out first on BraTS2021 dataset for preoperative segmentation. Performance is evaluated using DICE score (DS) and Hausdorff 95% (H95).ResultsIn preoperative scenario, overall DS is 91.09 (± 0.60) and H95 is 8.35 (± 1.12), considering tumor core, enhancing tumor and whole tumor (ET and edema). In postoperative context, overall DS is 72.31 (± 2.88) and H95 is 23.43 (± 7.24), considering resection cavity (RC), gross tumor volume (GTV) and whole tumor (WT). Remarkably, the RC segmentation obtained a mean DS of 63.52 (± 8.90) in postoperative MRIs.ConclusionsThe performances achieved by the algorithm are consistent with previous literature for both pre-operative and post-operative glioblastoma’s MRI evaluation. Through the proposed algorithm, it is possible to reduce the impact of low-quality images and missing sequences.

Raidionics: an open software for pre- and postoperative central nervous system tumor segmentation and standardized reporting

For patients suffering from central nervous system tumors, prognosis estimation, treatment decisions, and postoperative assessments are made from the analysis of a set of magnetic resonance (MR) scans. Currently, the lack of open tools for standardized and automatic tumor segmentation and generation of clinical reports, incorporating relevant tumor characteristics, leads to potential risks from inherent decisions’ subjectivity. To tackle this problem, the proposed Raidionics open-source software has been developed, offering both a user-friendly graphical user interface and stable processing backend. The software includes preoperative segmentation models for each of the most common tumor types (i.e., glioblastomas, lower grade gliomas, meningiomas, and metastases), together with one early postoperative glioblastoma segmentation model. Preoperative segmentation performances were quite homogeneous across the four different brain tumor types, with an average Dice around 85% and patient-wise recall and precision around 95%. Postoperatively, performances were lower with an average Dice of 41%. Overall, the generation of a standardized clinical report, including the tumor segmentation and features computation, requires about ten minutes on a regular laptop. The proposed Raidionics software is the first open solution enabling an easy use of state-of-the-art segmentation models for all major tumor types, including preoperative and postsurgical standardized reports.

An Automatic Deep Learning–Based Workflow for Glioblastoma Survival Prediction Using Preoperative Multimodal MR Images: A Feasibility Study

PurposeMost radiomic studies use the features extracted from the manually drawn tumor contours for classification or survival prediction. However, large interobserver segmentation variations lead to inconsistent features and hence introduce more challenges in constructing robust prediction models. Here, we proposed an automatic workflow for glioblastoma (GBM) survival prediction based on multimodal magnetic resonance (MR) images. Methods and MaterialsTwo hundred eighty-five patients with glioma (210 GBM, 75 low-grade glioma) were included. One hundred sixty-three of the patients with GBM had overall survival data. Every patient had 4 preoperative MR images and manually drawn tumor contours. A 3-dimensional convolutional neural network, VGG-Seg, was trained and validated using 122 patients with glioma for automatic GBM segmentation. The trained VGG-Seg was applied to the remaining 163 patients with GBM to generate their autosegmented tumor contours. The handcrafted and deep learning (DL)–based radiomic features were extracted from the autosegmented contours using explicitly designed algorithms and a pretrained convolutional neural network, respectively. One hundred sixty-three patients with GBM were randomly split into training (n = 122) and testing (n = 41) sets for survival analysis. Cox regression models were trained to construct the handcrafted and DL-based signatures. The prognostic powers of the 2 signatures were evaluated and compared. ResultsThe VGG-Seg achieved a mean Dice coefficient of 0.86 across 163 patients with GBM for GBM segmentation. The handcrafted signature achieved a C-index of 0.64 (95% confidence interval, 0.55-0.73), whereas the DL-based signature achieved a C-index of 0.67 (95% confidence interval, 0.57-0.77). Unlike the handcrafted signature, the DL-based signature successfully stratified testing patients into 2 prognostically distinct groups. ConclusionsThe VGG-Seg generated accurate GBM contours from 4 MR images. The DL-based signature achieved a numerically higher C-index than the handcrafted signature and significant patient stratification. The proposed automatic workflow demonstrated the potential of improving patient stratification and survival prediction in patients with GBM.

Challenges in Building of Deep Learning Models for Glioblastoma Segmentation: Evidence from Clinical Data.

In this article, we compare the performance of a state-of-the-art segmentation network (UNet) on two different glioblastoma (GB) segmentation datasets. Our experiments show that the same training procedure yields almost twice as bad results on the retrospective clinical data compared to the BraTS challenge data (in terms of Dice score). We discuss possible reasons for such an outcome, including inter-rater variability and high variability in magnetic resonance imaging (MRI) scanners and scanner settings. The high performance of segmentation models, demonstrated on preselected imaging data, does not bring the community closer to using these algorithms in clinical settings. We believe that a clinically applicable deep learning architecture requires a shift from unified datasets to heterogeneous data.

Radiomics may increase the prognostic value for survival in glioblastoma patients when combined with conventional clinical and genetic prognostic models.

To evaluate the additional prognostic value of multiparametric MR-based radiomics in patients with glioblastoma when combined with conventional clinical and genetic prognostic factors. In this single-center study, patients diagnosed with glioblastoma between October 2007 and December 2019 were retrospectively screened and grouped into training and test sets with a 7:3 distribution. Segmentations of glioblastoma using multiparametric MRI were performed automatically via a convolutional-neural network. Prognostic factors in the clinical model included age, sex, type of surgery/post-operative treatment, and tumor location; those in the genetic model included statuses of isocitrate dehydrogenase-1 mutation and O-6-methylguanine-DNA-methyltransferase promoter methylation. Univariate and multivariate Cox proportional hazards analyses were performed for overall survival (OS) and progression-free survival (PFS). Integrated time-dependent area under the curve (iAUC) for survival was calculated and compared between prognostic models via the bootstrapping method (performances were validated with prediction error curves). Overall, 120 patients were included (training set, 85; test set, 35). The mean OS and PFS were 25.5 and 18.6 months, respectively. The prognostic performances of multivariate models improved when radiomics was added to the clinical model (iAUC: OS, 0.62 to 0.73; PFS, 0.58 to 0.66), genetic model (iAUC: OS, 0.59 to 0.67; PFS, 0.59 to 0.65), and combined model (iAUC: OS, 0.65 to 0.73; PFS, 0.62 to 0.67). In the test set, the combined model (clinical, genetic, and radiomics) demonstrated robust validation for risk prediction of OS and PFS. Radiomics increased the prognostic value when combined with conventional clinical and genetic prognostic models for OS and PFS in glioblastoma patients. • CNN-based automatic segmentation of glioblastoma on multiparametric MRI was useful in extracting radiomic features. • Patients with glioblastoma with high-risk radiomics scores had poor overall survival (hazards ratio 8.33, p < 0.001) and progression-free survival (hazards ratio 3.76, p < 0.001). • MR-based radiomics improved the survival prediction when combined with clinical and genetic factors (overall and progression-free survival iAUC from 0.65 to 0.73 and 0.62 to 0.67, respectively; both p < 0.001).

Robust Deep Learning-based Segmentation of Glioblastoma on Routine Clinical MRI Scans Using Sparsified Training.

PurposeTo improve the robustness of deep learning–based glioblastoma segmentation in a clinical setting with sparsified datasets.Materials and MethodsIn this retrospective study, preoperative T1-weighted, T2-weighted, T2-weighted fluid-attenuated inversion recovery, and postcontrast T1-weighted MRI from 117 patients (median age, 64 years; interquartile range [IQR], 55–73 years; 76 men) included within the Multimodal Brain Tumor Image Segmentation (BraTS) dataset plus a clinical dataset (2012–2013) with similar imaging modalities of 634 patients (median age, 59 years; IQR, 49–69 years; 382 men) with glioblastoma from six hospitals were used. Expert tumor delineations on the postcontrast images were available, but for various clinical datasets, one or more sequences were missing. The convolutional neural network, DeepMedic, was trained on combinations of complete and incomplete data with and without site-specific data. Sparsified training was introduced, which randomly simulated missing sequences during training. The effects of sparsified training and center-specific training were tested using Wilcoxon signed rank tests for paired measurements.ResultsA model trained exclusively on BraTS data reached a median Dice score of 0.81 for segmentation on BraTS test data but only 0.49 on the clinical data. Sparsified training improved performance (adjusted P < .05), even when excluding test data with missing sequences, to median Dice score of 0.67. Inclusion of site-specific data during sparsified training led to higher model performance Dice scores greater than 0.8, on par with a model based on all complete and incomplete data. For the model using BraTS and clinical training data, inclusion of site-specific data or sparsified training was of no consequence.ConclusionAccurate and automatic segmentation of glioblastoma on clinical scans is feasible using a model based on large, heterogeneous, and partially incomplete datasets. Sparsified training may boost the performance of a smaller model based on public and site-specific data.Supplemental material is available for this article.Published under a CC BY 4.0 license.

Deep learning for glioblastoma segmentation using preoperative magnetic resonance imaging identifies volumetric features associated with survival

BackgroundMeasurement of volumetric features is challenging in glioblastoma. We investigate whether volumetric features derived from preoperative MRI using a convolutional neural network–assisted segmentation is correlated with survival.MethodsPreoperative MRI of 120 patients were scored using Visually Accessible Rembrandt Images (VASARI) features. We trained and tested a multilayer, multi-scale convolutional neural network on multimodal brain tumour segmentation challenge (BRATS) data, prior to testing on our dataset. The automated labels were manually edited to generate ground truth segmentations. Network performance for our data and BRATS data was compared. Multivariable Cox regression analysis corrected for multiple testing using the false discovery rate was performed to correlate clinical and imaging variables with overall survival.ResultsMedian Dice coefficients in our sample were (1) whole tumour 0.94 (IQR, 0.82–0.98) compared to 0.91 (IQR, 0.83–0.94 p = 0.012), (2) FLAIR region 0.84 (IQR, 0.63–0.95) compared to 0.81 (IQR, 0.69–0.8 p = 0.170), (3) contrast-enhancing region 0.91 (IQR, 0.74–0.98) compared to 0.83 (IQR, 0.78–0.89 p = 0.003) and (4) necrosis region were 0.82 (IQR, 0.47–0.97) compared to 0.67 (IQR, 0.42–0.81 p = 0.005). Contrast-enhancing region/tumour core ratio (HR 4.73 [95% CI, 1.67–13.40], corrected p = 0.017) and necrotic core/tumour core ratio (HR 8.13 [95% CI, 2.06–32.12], corrected p = 0.011) were independently associated with overall survival.ConclusionSemi-automated segmentation of glioblastoma using a convolutional neural network trained on independent data is robust when applied to routine clinical data. The segmented volumes have prognostic significance.

Multi-scale segmentation in GBM treatment using diffusion tensor imaging

Glioblastoma (GBM) is the commonest primary malignant brain tumor in adults, and despite advances in multi-modality therapy, the outlook for patients has changed little in the last 10 years. Local recurrence is the predominant pattern of treatment failure, hence improved local therapies (surgery and radiotherapy) are needed to improve patient outcomes. Currently segmentation of GBM for surgery or radiotherapy (RT) planning is labor intensive, especially for high-dimensional MR imaging methods that may provide more sensitive indicators of tumor phenotype. Automating processing and segmentation of these images will aid treatment planning. Diffusion tensor magnetic resonance imaging is a recently developed technique (DTI) that is exquisitely sensitive to the ordered diffusion of water in white matter tracts. Our group has shown that decomposition of the tensor information into the isotropic component (p – shown to represent tumor invasion) and the anisotropic component (q – shown to represent the tumor bulk) can provide valuable prognostic information regarding tumor infiltration and patient survival. However, tensor decomposition of DTI data is not commonly used for neurosurgery or radiotherapy treatment planning due to difficulties in segmenting the resultant image maps. For this reason, automated techniques for segmentation of tensor decomposition maps would have significant clinical utility. In this paper, we modified a well-established convolutional neural network architecture (CNN) for medical image segmentation and used it as an automatic multi-sequence GBM segmentation based on both DTI image maps (p and q maps) and conventional MRI sequences (T2-FLAIR and T1 weighted post contrast (T1c)). In this proof-of-concept work, we have used multiple MRI sequences, each with individually defined ground truths for better understanding of the contribution of each image sequence to the segmentation performance. The high accuracy and efficiency of our proposed model demonstrates the potential of utilizing diffusion MR images for target definition in precision radiation treatment planning and surgery in routine clinical practice.

Non-convex non-local reactive flows for saliency detection and segmentation

We propose and numerically solve a new variational model for automatic saliency detection and segmentation in digital images. Using a non-local framework we consider a family of edge preserving functions combined with a new quadratic saliency detection term. Such a term defines a constrained bilateral obstacle problem for image classification driven by p-Laplacian operators, including the so-called hyper-Laplacian case (0<p<1). As an application the related non-convex non-local reactive flows are considered for glioblastoma segmentation in magnetic resonance fluid-attenuated inversion recovery (MRI-Flair) images. A fast convolutional kernel based approximated solution is computed. The numerical experiments show that the non-convexity related to the hyper-Laplacian operators provokes sparseness of the non-local gradients and provides better results in terms of the standard metrics when the parameter p decreases.

Automatic Glioblastoma Segmentation in Multimodal MR Images Using Improved Fully Convolutional Neural Networks

Automatic Glioblastoma Segmentation in Multimodal MR Images Using Improved Fully Convolutional Neural Networks

Spatial logics and model checking for medical imaging

Recent research on spatial and spatio-temporal model checking provides novel image analysis methodologies, rooted in logical methods for topological spaces. Medical imaging (MI) is a field where such methods show potential for ground-breaking innovation. Our starting point is SLCS, the Spatial Logic for Closure Spaces—closure spaces being a generalisation of topological spaces, covering also discrete space structures—and topochecker, a model checker for SLCS (and extensions thereof). We introduce the logical language ImgQL (“Image Query Language”). ImgQL extends SLCS with logical operators describing distance and region similarity. The spatio-temporal model checker topochecker is correspondingly enhanced with state-of-the-art algorithms, borrowed from computational image processing, for efficient implementation of distance-based operators, namely distance transforms. Similarity between regions is defined by means of a statistical similarity operator, based on notions from statistical texture analysis. We illustrate our approach by means of an example of analysis of Magnetic Resonance images: segmentation of glioblastoma and its oedema.

Inter-rater agreement in glioma segmentations on longitudinal MRI.

BackgroundTumor segmentation of glioma on MRI is a technique to monitor, quantify and report disease progression. Manual MRI segmentation is the gold standard but very labor intensive. At present the quality of this gold standard is not known for different stages of the disease, and prior work has mainly focused on treatment-naive glioblastoma. In this paper we studied the inter-rater agreement of manual MRI segmentation of glioblastoma and WHO grade II-III glioma for novices and experts at three stages of disease. We also studied the impact of inter-observer variation on extent of resection and growth rate.MethodsIn 20 patients with WHO grade IV glioblastoma and 20 patients with WHO grade II-III glioma (defined as non-glioblastoma) both the enhancing and non-enhancing tumor elements were segmented on MRI, using specialized software, by four novices and four experts before surgery, after surgery and at time of tumor progression. We used the generalized conformity index (GCI) and the intra-class correlation coefficient (ICC) of tumor volume as main outcome measures for inter-rater agreement.ResultsFor glioblastoma, segmentations by experts and novices were comparable. The inter-rater agreement of enhancing tumor elements was excellent before surgery (GCI 0.79, ICC 0.99) poor after surgery (GCI 0.32, ICC 0.92), and good at progression (GCI 0.65, ICC 0.91). For non-glioblastoma, the inter-rater agreement was generally higher between experts than between novices. The inter-rater agreement was excellent between experts before surgery (GCI 0.77, ICC 0.92), was reasonable after surgery (GCI 0.48, ICC 0.84), and good at progression (GCI 0.60, ICC 0.80). The inter-rater agreement was good between novices before surgery (GCI 0.66, ICC 0.73), was poor after surgery (GCI 0.33, ICC 0.55), and poor at progression (GCI 0.36, ICC 0.73). Further analysis showed that the lower inter-rater agreement of segmentation on postoperative MRI could only partly be explained by the smaller volumes and fragmentation of residual tumor. The median interquartile range of extent of resection between raters was 8.3% and of growth rate was 0.22 mm/year.ConclusionManual tumor segmentations on MRI have reasonable agreement for use in spatial and volumetric analysis. Agreement in spatial overlap is of concern with segmentation after surgery for glioblastoma and with segmentation of non-glioblastoma by non-experts.

Clinical Evaluation of a Multiparametric Deep Learning Model for Glioblastoma Segmentation Using Heterogeneous Magnetic Resonance Imaging Data From Clinical Routine.

ObjectivesThe aims of this study were, first, to evaluate a deep learning–based, automatic glioblastoma (GB) tumor segmentation algorithm on clinical routine data from multiple centers and compare the results to a ground truth, manual expert segmentation, and second, to evaluate the quality of the segmentation results across heterogeneous acquisition protocols of routinely acquired clinical magnetic resonance imaging (MRI) examinations from multiple centers.Materials and MethodsThe data consisted of preoperative MRI scans (T1, T2, FLAIR, and contrast-enhanced [CE] T1) of 64 patients with an initial diagnosis of primary GB, which were acquired in 15 institutions with varying protocols. All images underwent preprocessing (coregistration, skull stripping, resampling to isotropic resolution, normalization) and were fed into an independently trained deep learning model based on DeepMedic, a multilayer, multiscale convolutional neural network for detection and segmentation of tumor compartments. Automatic segmentation results for the whole tumor, necrosis, and CE tumor were compared with manual segmentations.ResultsWhole tumor and CE tumor compartments were correctly detected in 100% of the cases; necrosis was correctly detected in 91% of the cases. A high segmentation accuracy comparable to interrater variability was achieved for the whole tumor (mean dice similarity coefficient [DSC], 0.86 ± 0.09) and CE tumor (DSC, 0.78 ± 0.15). The DSC for tumor necrosis was 0.62 ± 0.30. We have observed robust segmentation quality over heterogeneous image acquisition protocols, for example, there were no correlations between resolution and segmentation accuracy of the single tumor compartments. Furthermore, no relevant correlation was found between quality of automatic segmentation and volume of interest properties (surface-to-volume ratio and volume).ConclusionsThe proposed approach for automatic segmentation of GB proved to be robust on routine clinical data and showed on all tumor compartments a high automatic detection rate and a high accuracy, comparable to interrater variability. Further work on improvements of the segmentation accuracy for the necrosis compartments should be guided by the evaluation of the clinical relevance.Therefore, we propose this approach as a suitable building block for automatic tumor segmentation to support radiologists or neurosurgeons in the preoperative reading of GB MRI images and characterization of primary GB.

Effective user interaction in online interactive semantic segmentation of glioblastoma magnetic resonance imaging.

Interactive segmentation is a promising approach to solving the pervasive shortage of reference annotations for automated medical image processing. We focus on the challenging task of glioblastoma segmentation in magnetic resonance imaging using a random forest pixel classifier trained iteratively on scribble annotations. Our experiments use data from the MICCAI Multimodal Brain Tumor Segmentation Challenge 2013 and simulate expert interactions using different approaches: corrective annotations, class-balanced corrections, annotations where classifier uncertainty is high, and corrections where classifier uncertainty is high/low. We find that it is better to correct the classifier than to provide annotations where the classifier is uncertain, resulting in significantly better Dice scores in the edema (0.662 to 0.686) and necrosis (0.550 to 0.676) regions after 20 interactions. It is also advantageous to balance inputs among classes, with significantly better Dice in the necrotic (0.501 to 0.676) and nonenhancing (0.151 to 0.235) regions compared to fully random corrections. Corrective annotations in regions of high classifier uncertainty provide no additional benefit, low uncertainty corrections perform worst. Preliminary experiments with real users indicate that those with intermediate proficiency make a considerable number of annotation errors. The performance of corrective approaches suffers most strongly from this, leading to a less profound difference to uncertainty-based annotations.