Search

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) and ultrasound (US), which are two common modalities for clinical breast tumor diagnosis besides Mammograms, can provide different and complementary information for the same tumor regions. Although many machine learning methods have been proposed for breast tumor classification based on either single modality, it remains unclear how to further boost the classification performance by utilizing paired multi-modality information with different dimensions. In this paper, we propose MRI-US multi-modality network (MUM-Net) to classify breast tumor into different subtypes based on 3D MR and 2D US images. The key insight of MUM-Net is that we explicitly distill modality-agnostic features for tumor classification. Specifically, we first adopt a discrimination-adaption module to decompose features into modality-agnostic and modality-specific ones with min-max training strategies. Then, we propose a feature fusion module to increase the compactness of the modality-agnostic features by utilizing an affinity matrix with nearest neighbour selection. We build a paired MRI-US breast tumor classification dataset containing 502 cases with three clinical indicators to validate the proposed method. In three tasks including lymph node metastasis, histological grade and Ki-67 level, MUM-Net achieves AUC scores of 0.8581, 0.8965 and 0.8577, outperforming other counterparts which are based on single task or single modality by a wide margin. In addition, we find that the extracted modality-agnostic features can help the network focus on the tumor regions in both modalities.

Deep learning models usually suffer from domain shift issues, where models trained on one source domain do not generalize well to other unseen domains. In this work, we investigate the single-source domain generalization problem: training a deep network that is robust to unseen domains, under the condition that training data is only available from one source domain, which is common in medical imaging applications. We tackle this problem in the context of cross-domain medical image segmentation. Under this scenario, domain shifts are mainly caused by different acquisition processes. We propose a simple causality-inspired data augmentation approach to expose a segmentation model to synthesized domain-shifted training examples. Specifically, 1) to make the deep model robust to discrepancies in image intensities and textures, we employ a family of randomly-weighted shallow networks. They augment training images using diverse appearance transformations. 2) Further we show that spurious correlations among objects in an image are detrimental to domain robustness. These correlations might be taken by the network as domain-specific clues for making predictions, and they may break on unseen domains. We remove these spurious correlations via causal intervention. This is achieved by resampling the appearances of potentially correlated objects independently. The proposed approach is validated on three cross-domain segmentation tasks: cross-modality (CT-MRI) abdominal image segmentation, cross-sequence (bSSFP-LGE) cardiac MRI segmentation, and cross-center prostate MRI segmentation. The proposed approach yields consistent performance gains compared with competitive methods when tested on unseen domains., This is an early, non-peer-reviewed version. For the final peer-reviewed full version that has been substantially revised, please find: https://ieeexplore.ieee.org/document/9961940. Please find the code at https://github.com/cheng-01037/Causality-Medical-Image-Domain-Generalization

Radiomics has potential advantages in the noninvasive histopathological and molecular diagnosis of gliomas. We aimed to develop a novel image signature (IS)-based radiomics model to achieve multilayered preoperative diagnosis and prognostic stratification of gliomas. Herein, we established three separate case cohorts, consisting of 655 glioma patients, and carried out a retrospective study. Image and clinical data of three cohorts were used for training (N = 188), cross-validation (N = 411), and independent testing (N = 56) of the IS model. All tumors were segmented from magnetic resonance (MR) images by the 3D U-net, followed by extraction of high-throughput network features, which were referred to as IS. IS was then used to perform noninvasive histopathological diagnosis and molecular subtyping. Moreover, a new IS-based clustering method was applied for prognostic stratification in IDH-wild-type lower-grade glioma (IDHwt LGG) and triple-negative glioblastoma (1p19q retain/IDH wild-type/TERTp-wild-type GBM). The average accuracies of histological diagnosis and molecular subtyping were 89.8 and 86.1% in the cross-validation cohort, while these numbers reached 83.9 and 80.4% in the independent testing cohort. IS-based clustering method was demonstrated to successfully divide IDHwt LGG into two subgroups with distinct median overall survival time (48.63 vs 38.27 months respectively, P = 0.023), and two subgroups in triple-negative GBM with different median OS time (36.8 vs 18.2 months respectively, P = 0.013). Our findings demonstrate that our novel IS-based radiomics model is an effective tool to achieve noninvasive histo-molecular pathological diagnosis and prognostic stratification of gliomas. This IS model shows potential for future routine use in clinical practice.

Data privacy mechanisms are essential for rapidly scaling medical training databases to capture the heterogeneity of patient data distributions toward robust and generalizable machine learning systems. In the current COVID-19 pandemic, a major focus of artificial intelligence (AI) is interpreting chest CT, which can be readily used in the assessment and management of the disease. This paper demonstrates the feasibility of a federated learning method for detecting COVID-19 related CT abnormalities with external validation on patients from a multinational study. We recruited 132 patients from seven multinational different centers, with three internal hospitals from Hong Kong for training and testing, and four external, independent datasets from Mainland China and Germany, for validating model generalizability. We also conducted case studies on longitudinal scans for automated estimation of lesion burden for hospitalized COVID-19 patients. We explore the federated learning algorithms to develop a privacy-preserving AI model for COVID-19 medical image diagnosis with good generalization capability on unseen multinational datasets. Federated learning could provide an effective mechanism during pandemics to rapidly develop clinically useful AI across institutions and countries overcoming the burden of central aggregation of large amounts of sensitive data.

Machine learning models fail to perform well on real-world applications when 1) the category distribution P(Y) of the training dataset suffers from long-tailed distribution and 2) the test data is drawn from different conditional distributions P(X|Y). Existing approaches cannot handle the scenario where both issues exist, which however is common for real-world applications. In this study, we took a step forward and looked into the problem of long-tailed classification under domain shifts. We designed three novel core functional blocks including Distribution Calibrated Classification Loss, Visual-Semantic Mapping and Semantic-Similarity Guided Augmentation. Furthermore, we adopted a meta-learning framework which integrates these three blocks to improve domain generalization on unseen target domains. Two new datasets were proposed for this problem, named AWA2-LTS and ImageNet-LTS. We evaluated our method on the two datasets and extensive experimental results demonstrate that our proposed method can achieve superior performance over state-of-the-art long-tailed/domain generalization approaches and the combinations. Source codes and datasets can be found at our project page https://xiaogu.site/LTDS., Comment: accepted to ECCV 2022

Image registration is a fundamental medical image analysis task, and a wide variety of approaches have been proposed. However, only a few studies have comprehensively compared medical image registration ap- proaches on a wide range of clinically relevant tasks. This limits the development of registration methods, the adoption of research advances into practice, and a fair benchmark across competing approaches. The Learn2Reg challenge addresses these limitations by providing a multi- task medical image registration data set for comprehen- sive characterisation of deformable registration algorithms. A continuous evaluation will be possible at https:// learn2reg.grand-challenge.org. Learn2Reg covers a wide range of anatomies (brain, abdomen, and thorax), modalities (ultrasound, CT, MR), availability of annotations, as well as intra- and inter-patient registration evaluation. We established an easily accessible framework for training and validation of 3D registration methods, which enabled the compilation of results of over 65 individual method submissions from more than 20 unique teams. We used a complementary set of metrics, including robustness, ac- curacy, plausibility, and runtime, enabling unique insight into the current state-of-the-art of medical image regis- tration. This paper describes datasets, tasks, evaluation methods and results of the challenge, as well as results of further analysis of transferability to new datasets, the importance of label supervision, and resulting bias. While no single approach worked best across all tasks, many methodological aspects could be identified that push the performance of medical image registration to new state- of-the-art performance. Furthermore, we demystified the common belief that conventional registration methods have to be much slower than deep-learning-based methods.

Early identification of stroke is critical for the treatment and subsequent recovery. Non-contrast computed tomography (ncCT) is a routinely employed imaging modality for stroke evaluation. However, the early identification of stroke in ncCT images is very difficult, since there are subtle differences between lesion and healthy tissue during the hyperacute phase. In this paper, an image patch classification-based method was developed to detect the early ischemic stroke that is invisible to the radiologist in ncCT. First, we proposed radiomics-based patch classification model to identify whether each patch in ncCT is stroke region or not. To improve the identification accuracy, a symmetry image patch classification was developed, in which an image patch in one brain hemisphere was combined with its contralateral image patch for classification. Second, based on the spatial dependencies between neighboring patches, we built a maximum a posteriori identification model which integrated the spatial constraint information to improve the identification performance. Finally, after performing morphological post-processing, we defined the detected results containing more than 300 voxels as the stroke region. 108 stroke cases that were all invisible to radiologists in ncCT were enrolled in the study and divided into one training cohort and two independent testing cohorts to validate the proposed method. The proposed method has achieved identification accuracies of 76.67% and 75.00% on the two independent testing cohorts, respectively. The results proved the potential of the proposed radiomics model in the task which possessed great clinical values.

Due to their high spatial resolution, thin-section magnetic resonance (MR) images serve as ideal medical images for brain structure investigation and brain surgery navigation. However, compared with the clinically widely used thick-section MR images, thin-section MR images are less available due to the imaging cost. Thin-section MR images of infants are even scarcer but are quite valuable for the study of human brain development. Therefore, we propose a method for the reconstruction of thin-section MR images from thick-section images. A two-stage reconstruction framework based on generative adversarial networks (GANs) and a convolutional neural network (CNN) is proposed to reconstruct thin-section MR images from thick-section images in the axial and sagittal planes. A 3D-Y-Net-GAN is first proposed to fuse MR images from the axial and sagittal planes and to achieve the first-stage thin-section reconstruction. A 3D-DenseU-Net followed by a stack of enhanced residual blocks is then proposed to provide further detail recalibrations and structural corrections in the sagittal plane. In this method, a comprehensive loss function is also proposed to help the networks capture more structural details. The reconstruction performance of the proposed method is compared with bicubic interpolation, sparse representation, and 3D-SRU-Net. Cross-validation based on 35 cases and independent testing based on two datasets with totally 114 cases reveal that, compared with the other three methods, the proposed method provides an average 23.5% improvement in peak signal-to-noise ratio (PSNR), 90.5% improvement in structural similarity (SSIM), and 21.5% improvement in normalized mutual information (NMI). The quantitative evaluation and visual inspection demonstrate that our proposed method outperforms those methods by reconstructing more realistic results with better structural details.

The success of neural networks on medical image segmentation tasks typically relies on large labeled datasets for model training. However, acquiring and manually labeling a large medical image set is resource-intensive, expensive, and sometimes impractical due to data sharing and privacy issues. To address this challenge, we propose AdvChain, a generic adversarial data augmentation framework, aiming at improving both the diversity and effectiveness of training data for medical image segmentation tasks. AdvChain augments data with dynamic data augmentation, generating randomly chained photo-metric and geometric transformations to resemble realistic yet challenging imaging variations to expand training data. By jointly optimizing the data augmentation model and a segmentation network during training, challenging examples are generated to enhance network generalizability for the downstream task. The proposed adversarial data augmentation does not rely on generative networks and can be used as a plug-in module in general segmentation networks. It is computationally efficient and applicable for both low-shot supervised and semi-supervised learning. We analyze and evaluate the method on two MR image segmentation tasks: cardiac segmentation and prostate segmentation with limited labeled data. Results show that the proposed approach can alleviate the need for labeled data while improving model generalization ability, indicating its practical value in medical imaging applications., Comment: Under review

Radiomics has potential advantages in the noninvasive histopathological and molecular diagnosis of gliomas. We aimed to develop a novel image signature (IS)-based radiomics model to achieve multilayered preoperative diagnosis and prognostic stratification of gliomas. Herein, we established three separate case cohorts, consisting of 655 glioma patients, and carried out a retrospective study. Image and clinical data of three cohorts were used for training (N = 188), cross-validation (N = 411), and independent testing (N = 56) of the IS model. All tumors were segmented from magnetic resonance (MR) images by the 3D U-net, followed by extraction of high-throughput network features, which were referred to as IS. IS was then used to perform noninvasive histopathological diagnosis and molecular subtyping. Moreover, a new IS-based clustering method was applied for prognostic stratification in IDH-wild-type lower-grade glioma (IDHwt LGG) and triple-negative glioblastoma (1p19q retain/IDH wild-type/TERTp-wild-type GBM). The average accuracies of histological diagnosis and molecular subtyping were 89.8 and 86.1% in the cross-validation cohort, while these numbers reached 83.9 and 80.4% in the independent testing cohort. IS-based clustering method was demonstrated to successfully divide IDHwt LGG into two subgroups with distinct median overall survival time (48.63 vs 38.27 months respectively, P = 0.023), and two subgroups in triple-negative GBM with different median OS time (36.8 vs 18.2 months respectively, P = 0.013). Our findings demonstrate that our novel IS-based radiomics model is an effective tool to achieve noninvasive histo-molecular pathological diagnosis and prognostic stratification of gliomas. This IS model shows potential for future routine use in clinical practice.

The status of isocitrate dehydrogenase 1 (IDH1) is highly correlated with the development, treatment and prognosis of glioma. We explored a noninvasive method to reveal IDH1 status by using a quantitative radiomics approach for grade II glioma.A primary cohort consisting of 110 patients pathologically diagnosed with grade II glioma was retrospectively studied. The radiomics method developed in this paper includes image segmentation, high-throughput feature extraction, radiomics sequencing, feature selection and classification. Using the leave-one-out cross-validation (LOOCV) method, the classification result was compared with the real IDH1 situation from Sanger sequencing. Another independent validation cohort containing 30 patients was utilised to further test the method.A total of 671 high-throughput features were extracted and quantized. 110 features were selected by improved genetic algorithm. In LOOCV, the noninvasive IDH1 status estimation based on the proposed approach presented an estimation accuracy of 0.80, sensitivity of 0.83 and specificity of 0.74. Area under the receiver operating characteristic curve reached 0.86. Further validation on the independent cohort of 30 patients produced similar results.Radiomics is a potentially useful approach for estimating IDH1 mutation status noninvasively using conventional T2-FLAIR MRI images. The estimation accuracy could potentially be improved by using multiple imaging modalities.• Noninvasive IDH1 status estimation can be obtained with a radiomics approach. • Automatic and quantitative processes were established for noninvasive biomarker estimation. • High-throughput MRI features are highly correlated to IDH1 states. • Area under the ROC curve of the proposed estimation method reached 0.86.

Thin-section magnetic resonance imaging (MRI) can provide higher resolution anatomical structures and more precise clinical information than thick-section images. However, thin-section MRI is not always available due to the imaging cost issue. In multicenter retrospective studies, a large number of data are often in thick-section manner with different section thickness. The lack of thin-section data and the difference in section thickness bring considerable difficulties in the study based on the image big data. In this article, we introduce DeepVolume, a two-step deep learning architecture to address the challenge of accurate thin-section MR image reconstruction. The first stage is the brain structure-aware network, in which the thick-section MR images in axial and sagittal planes are fused by a multitask 3-D U-net with prior knowledge of brain volume segmentation, which encourages the reconstruction result to have correct brain structure. The second stage is the spatial connection-aware network, in which the preliminary reconstruction results are adjusted slice-by-slice by a recurrent convolutional network embedding convolutional long short-term memory (LSTM) block, which enhances the precision of the reconstruction by utilizing the previously unassessed sagittal information. We used 305 paired brain MRI samples with thickness of 1.0 mm and 6.5 mm in this article. Extensive experiments illustrate that DeepVolume can produce the state-of-the-art reconstruction results by embedding more anatomical knowledge. Furthermore, considering DeepVolume as an intermediate step, the practical and clinical value of our method is validated by applying the brain volume estimation and voxel-based morphometry. The results show that DeepVolume can provide much more reliable brain volume estimation in the normalized space based on the thick-section MR images compared with the traditional solutions.

Although chest X-ray (CXR) offers a 2D projection with overlapped anatomies, it is widely used for clinical diagnosis. There is clinical evidence supporting that decomposing an X-ray image into different components (e.g., bone, lung and soft tissue) improves diagnostic value. We hereby propose a decomposition generative adversarial network (DecGAN) to anatomically decompose a CXR image but with unpaired data. We leverage the anatomy knowledge embedded in CT, which features a 3D volume with clearly visible anatomies. Our key idea is to embed CT priori decomposition knowledge into the latent space of unpaired CXR autoencoder. Specifically, we train DecGAN with a decomposition loss, adversarial losses, cycle-consistency losses and a mask loss to guarantee that the decomposed results of the latent space preserve realistic body structures. Extensive experiments demonstrate that DecGAN provides superior unsupervised CXR bone suppression results and the feasibility of modulating CXR components by latent space disentanglement. Furthermore, we illustrate the diagnostic value of DecGAN and demonstrate that it outperforms the state-of-the-art approaches in terms of predicting 11 out of 14 common lung diseases.

It is of great clinical significance in the differential diagnosis of primary central nervous system lymphoma (PCNSL) and glioblastoma (GBM) because there are enormous differences between them in terms of therapeutic regimens. In this paper, we propose a system based on sparse representation for automatic classification of PCNSL and GBM. The proposed system distinguishes the two tumors by using of the different texture detail information of the two tumors on T1 contrast magnetic resonance imaging (MRI) images. First, inspired by the process of radiomics, we designed a dictionary learning and sparse representation-based method to extract texture information, and with this approach, the tumors with different volume and shape were transformed into 968 quantitative texture features. Next, aiming at the problem of the redundancy in the extracted features, feature selection based on iterative sparse representation was set up to select some key texture features with high stability and discrimination. Finally, the selected key features are used for differentiation based on sparse representation classification (SRC) method. By using ten-fold cross-validation method, the differentiation based on the proposed approach presents accuracy of 96.36%, sensitivity 96.30%, and specificity 96.43%. Experimental results show that our approach not only effectively distinguish the two tumors but also has strong robustness in practical application since it avoids the process of parameter extraction on advanced MRI images.临床上原发性脑部淋巴瘤（PCNSL）和胶质母细胞瘤（GBM）的治疗方案存在很大差异，因此治疗前对二者的精确鉴别具有重要临床价值。本文提出一套基于稀疏表示体系的肿瘤自动鉴别方法，利用 PCNSL 和 GBM T1 加权磁共振成像（MRI）图像纹理细节信息的差异鉴别这两种肿瘤。首先，基于影像组学的思想，设计一种基于字典学习和稀疏表示的肿瘤纹理特征提取方法，将不同体积、不同形状的肿瘤区域转化为 968 维纹理特征；其次，针对提取特征存在的冗余问题，建立迭代稀疏表示方法选择少数高稳定性高分辨力的特征；最后，将选择的关键特征送入稀疏表示分类器（SRC）分类。利用十折法对数据集进行交叉验证，鉴别结果的准确率为 96.36%，敏感度为 96.30%，特异性为 96.43%。实验结果表明，本文方法不仅能够有效地鉴别 PCNSL 和 GBM，还避免了使用先进 MRI 鉴别肿瘤时存在的参数提取问题，在实际应用中具有较强的鲁棒性。.

Cardiac Magnetic Resonance (CMR) imaging is widely used in the clinic to assess the patient-specific cardiac structure and function. However, the manual analysis of the CMR data is tedious and subjective. In this work, we developed a fully automatic segmentation system for the left ventricle (LV) myocardium from MR cine images. The system consists of three major components. Firstly, a conventional convolutional neural network (CNN) was trained to detect the global region of interest (ROI) of LV. Secondly, a novel multi-scope CNN was proposed to segment the LV myocardium from the reduced ROI, taking advantage of the image context in different scopes, such that both the local accuracy and global consistency can be implicitly learned by the CNN. Finally the results were pruned with simple morphological filtering preserving the largest component. With a relatively small training set of 200 MR cine images, the method achieved an average segmentation accuracy of 0.71 as expressed by the Dice overlap index. The proposed method can be applied to automatically segment the LV from MR images with the reasonable accuracy, or as a proper initialization for local shape methods to achieve further refined results.

Recently, the convolutional neural network (CNN) has been successfully applied to the task of brain tumor segmentation. However, the effectiveness of a CNN-based method is limited by the small receptive field, and the segmentation results don’t perform well in the spatial contiguity. Therefore, many attempts have been made to strengthen the spatial contiguity of the network output. In this paper, we proposed an adversarial training approach to train the CNN network. A discriminator network is trained along with a generator network which produces the synthetic segmentation results. The discriminator network is encouraged to discriminate the synthetic labels from the ground truth labels. Adversarial adjustments provided by the discriminator network are fed back to the generator network to help reduce the differences between the synthetic labels and the ground truth labels and reinforce the spatial contiguity with high-order loss terms. The presented method is evaluated on the Brats2017 training dataset. The experiment results demonstrate that the presented method could enhance the spatial contiguity of the segmentation results and improve the segmentation accuracy.

Due to the totally different therapeutic regimens needed for primary central nervous system lymphoma (PCNSL) and glioblastoma (GBM), accurate differentiation of the two diseases by noninvasive imaging techniques is important for clinical decision-making.Thirty cases of PCNSL and 66 cases of GBM with conventional T1-contrast magnetic resonance imaging (MRI) were analyzed in this study. Convolutional neural networks was used to segment tumor automatically. A modified scale invariant feature transform (SIFT) method was utilized to extract three-dimensional local voxel arrangement information from segmented tumors. Fisher vector was proposed to normalize the dimension of SIFT features. An improved genetic algorithm (GA) was used to extract SIFT features with PCNSL and GBM discrimination ability. The data-set was divided into a cross-validation cohort and an independent validation cohort by the ratio of 2:1. Support vector machine with the leave-one-out cross-validation based on 20 cases of PCNSL and 44 cases of GBM was employed to build and validate the differentiation model.Among 16,384 high-throughput features, 1356 features show significant differences between PCNSL and GBM with p0.05 and 420 features with p0.001. A total of 496 features were finally chosen by improved GA algorithm. The proposed method produces PCNSL vs. GBM differentiation with an area under the curve (AUC) curve of 99.1% (98.2%), accuracy 95.3% (90.6%), sensitivity 85.0% (80.0%) and specificity 100% (95.5%) on the cross-validation cohort (and independent validation cohort).Since the local voxel arrangement characterization provided by SIFT features, proposed method produced more competitive PCNSL and GBM differentiation performance by using conventional MRI than methods based on advanced MRI.

Glioblastoma is the most aggressive malignant brain tumor with poor prognosis. Radiomics is a newly emerging and promising technique to reveal the complex relationships between high-throughput medical image features and deep information of disease including pathology, biomarkers and genomics. An approach was developed to investigate the internal relationship between magnetic resonance imaging (MRI) features and the age-related origins of glioblastomas based on a quantitative radiomics method. A fully automatic image segmentation method was applied to segment the tumor regions from three dimensional MRI images. 555 features were then extracted from the image data. By analyzing large numbers of quantitative image features, some predictive and prognostic information could be obtained by the radiomics approach. 96 patients diagnosed with glioblastoma pathologically have been divided into two age groups (45 and ≥45 years old). As expected, there are 101 features showing the consistency with the age groups (T test, p .05), and unsupervised clustering results of those features also show coherence with the age difference (T test, p= .006). In conclusion, glioblastoma in different age groups present different radiomics-feature patterns with statistical significance, which indicates that glioblastoma in different age groups should have different pathologic, protein, or genic origins.

Deep learning-based radiomics (DLR) was developed to extract deep information from multiple modalities of magnetic resonance (MR) images. The performance of DLR for predicting the mutation status of isocitrate dehydrogenase 1 (IDH1) was validated in a dataset of 151 patients with low-grade glioma. A modified convolutional neural network (CNN) structure with 6 convolutional layers and a fully connected layer with 4096 neurons was used to segment tumors. Instead of calculating image features from segmented images, as typically performed for normal radiomics approaches, image features were obtained by normalizing the information of the last convolutional layers of the CNN. Fisher vector was used to encode the CNN features from image slices of different sizes. High-throughput features with dimensionality greater than 1.6*104 were obtained from the CNN. Paired t-tests and F-scores were used to select CNN features that were able to discriminate IDH1. With the same dataset, the area under the operating characteristic curve (AUC) of the normal radiomics method was 86% for IDH1 estimation, whereas for DLR the AUC was 92%. The AUC of IDH1 estimation was further improved to 95% using DLR based on multiple-modality MR images. DLR could be a powerful way to extract deep information from medical images.

This work proposed a novel automatic three-dimensional (3D) magnetic resonance imaging (MRI) segmentation method which would be widely used in the clinical diagnosis of the most common and aggressive brain tumor, namely, glioma. The method combined a multipathway convolutional neural network (CNN) and fully connected conditional random field (CRF). Firstly, 3D information was introduced into the CNN which makes more accurate recognition of glioma with low contrast. Then, fully connected CRF was added as a postprocessing step which purposed more delicate delineation of glioma boundary. The method was applied to T2flair MRI images of 160 low-grade glioma patients. With 59 cases of data training and manual segmentation as the ground truth, the Dice similarity coefficient (DSC) of our method was 0.85 for the test set of 101 MRI images. The results of our method were better than those of another state-of-the-art CNN method, which gained the DSC of 0.76 for the same dataset. It proved that our method could produce better results for the segmentation of low-grade gliomas.