Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries : 6th International Workshop, BrainLes 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 4, 2020, Revised Selected Papers, Part II

This two-volume set LNCS 12658 and 12659 constitutes the thoroughly refereed proceedings of the 6th International MICCAI Brainlesion Workshop, BrainLes 2020, the International Multimodal Brain Tumor Segmentation (BraTS) challenge, and the Computational Precision Medicine: Radiology-Pathology Challenge on Brain Tumor Classification (CPM-RadPath) challenge. These were held jointly at the 23rd Medical Image Computing for Computer Assisted Intervention Conference, MICCAI 2020, in Lima, Peru, in October 2020.\* The revised selected papers presented in these volumes were organized in the following topical sections: brain lesion image analysis (16 selected papers from 21 submissions); brain tumor image segmentation (69 selected papers from 75 submissions); and computational precision medicine: radiology-pathology challenge on brain tumor classification (6 selected papers from 6 submissions). \*The workshop and challenges were held virtually. Preface Organization Contents – Part II Contents – Part I Brain Tumor Segmentation Lightweight U-Nets for Brain Tumor Segmentation 1 Introduction 2 Data 3 Methods 3.1 Data Pre-processing 3.2 Our Deep Network Architecture 4 Experimental Study 4.1 Setup 4.2 The Results 5 Conclusions References Efficient Brain Tumour Segmentation Using Co-registered Data and Ensembles of Specialised Learners 1 Introduction 2 Related Work 2.1 BraTS: Challenge and Data Set 2.2 Related Literature 3 Methodology 3.1 Data Pre-processing 3.2 Model Architecture 4 Empirical Evaluation 4.1 Results 4.2 Performance Comparison 4.3 Analysis 4.4 Impact and Implications of Loss Functions 4.5 Performance Evaluation 5 Discussion 6 Conclusions and Future Work References Efficient MRI Brain Tumor Segmentation Using Multi-resolution Encoder-Decoder Networks 1 Introduction 2 Materials and Methods 2.1 Dataset 2.2 Segmentation 2.3 Survival Prediction 3 Results 3.1 Segmentation Task 3.2 Survival Prediction 4 Conclusion References Trialing U-Net Training Modifications for Segmenting Gliomas Using Open Source Deep Learning Framework 1 Introduction 2 Methods 2.1 Data 2.2 Model Architecture 2.3 Augmentation 2.4 Training 3 Experiments 3.1 Effects of Thresholding 3.2 Contours, Permutations, and Grouped Labels 3.3 Ensembles 3.4 Test Set 4 Results 4.1 Effects of Thresholding 4.2 Contours, Permutations, and Grouped Labels 4.3 Ensembles 4.4 Test Set Results 5 Discussion 6 Conclusion References HI-Net: Hyperdense Inception 3D UNet for Brain Tumor Segmentation 1 Introduction 2 Proposed Method 3 Implementation Details 3.1 Dataset 3.2 Experiments 3.3 Evaluation Metrics 3.4 Results 4 Conclusion References H2NF-Net for Brain Tumor Segmentation Using Multimodal MR Imaging: 2nd Place Solution to BraTS Challenge 2020 Segmentation Task 1 Introduction 2 Dataset 3 Method 3.1 Single HNF-Net 3.2 Cascaded HNF-Net 4 Experiments and Results 4.1 Implementation Details 4.2 Results on the BraTS 2020 Challenge Dataset 5 Conclusion References 2D Dense-UNet: A Clinically Valid Approach to Automated Glioma Segmentation 1 Introduction 2 Methods 3 Results 4 Discussion 5 Conclusion References Attention U-Net with Dimension-Hybridized Fast Data Density Functional Theory for Automatic Brain Tumor Image Segmentation 1 Introduction 2 Methods 2.1 Normalization and Augmentation 2.2 Feature Pre-extraction Using Fast Data Density Functional Theory 2.3 Encoder, Decoder, and Attention Block 2.4 Optimization 3 Results 4 Conclusion References MVP U-Net: Multi-View Pointwise U-Net for Brain Tumor Segmentation 1 Introduction 2 Methods 2.1 Preprocessing 2.2 Network Architecture 2.3 Loss 2.4 Optimization 3 Experiments and Results 4 Conclusion References Glioma Segmentation with 3D U-Net Backed with Energy-Based Post-Processing 1 Introduction 1.1 Related Work 2 Data 3 Methods 3.1 Preprocessing 3.2 Model 3.3 Loss Function 3.4 Training 3.5 Post Processing 4 Performance Evaluation 5 Conclusion References nnU-Net for Brain Tumor Segmentation 1 Introduction 2 Method 2.1 Rankings Should Be Used for Model Selection 2.2 nnU-Net Baseline 2.3 BraTS-Specific Optimizations 2.4 Further nnU-Net Modifications 3 Results 3.1 Aggregated Scores 3.2 Internal BraTS-Like Ranking 3.3 Qualitative Results 3.4 Test Set Results 4 Discussion References A Deep Random Forest Approach for Multimodal Brain Tumor Segmentation 1 Introduction 2 Proposed Architecture 3 Experimental Results 3.1 Dataset 3.2 Preprocessing 3.3 Feature Generation 3.4 Implementation Details 3.5 Performance 4 Conclusion References Brain Tumor Segmentation and Associated Uncertainty Evaluation Using Multi-sequences MRI Mixture Data Preprocessing 1 Introduction 2 Datasets Description 2.1 BraTS Dataset 2.2 Siberian Brain Tumor Dataset 3 Methods 4 Results 5 Conclusions References A Deep Supervision CNN Network for Brain Tumor Segmentation 1 Introduction 2 Related Work 3 Methods 3.1 Main Network 3.2 Deep Supervision Method 3.3 Loss Functions 4 Experiments 4.1 Pre-processing 4.2 Post-processing 4.3 Training Details 5 Results 6 Conclusion References Multi-threshold Attention U-Net (MTAU) Based Model for Multimodal Brain Tumor Segmentation in MRI Scans 1 Introduction 2 Methods 2.1 Dataset 2.2 Preprocessing 2.3 Model Architecture 3 Figures of Merit 3.1 Dice Coefficient (DSC) 3.2 Sensitivity (SN) 3.3 Specificity (SP) 3.4 Hausdorff Distance (h) 4 Results and Discussions 5 Conclusions References Multi-stage Deep Layer Aggregation for Brain Tumor Segmentation 1 Introduction 2 Methods 2.1 Deep Layer Aggregation 2.2 Loss Function 3 Experimental Setup 3.1 Data 3.2 Pre-processing and Data Augmentation 3.3 Settings and Model Training 3.4 Post-processing 4 Results 5 Conclusion References Glioma Segmentation Using Ensemble of 2D/3D U-Nets and Survival Prediction Using Multiple Features Fusion 1 Introduction 2 Proposed Methodology 2.1 Segmentation Task 2.2 Survival Prediction Task 3 Experiments 3.1 Dataset 3.2 Implementation Details 4 Results 5 Conclusion References Generalized Wasserstein Dice Score, Distributionally Robust Deep Learning, and Ranger for Brain Tumor Segmentation: BraTS 2020 Challenge 1 Introduction 2 Method: Varying the Three Main Ingredients of the Optimization of Deep Neural Networks 2.1 Changing the Per-Sample Loss Function: The Generalized Wasserstein Dice Loss ch18fidon2017generalised 2.2 Changing the Optimization Problem: Distributionally Robust Optimization ch18fidon2020sgd 2.3 Changing the Optimizer: Ranger ch18liu2019variance,ch18zhang2019lookahead 2.4 Deep Neural Networks Ensembling 3 Experiments and Results 3.1 Data and Implementation Details 3.2 Models Description 3.3 Mean Segmentation Performance 3.4 Robustness Performance 4 Conclusion References 3D Semantic Segmentation of Brain Tumor for Overall Survival Prediction 1 Introduction 1.1 Literature Review: BraTS 2019 2 Dataset 3 Proposed Method 3.1 Task 1: Tumor Segmentation 3.2 Task 2: Overall Survival Prediction 4 Implementation Details 4.1 Pre-processing 4.2 Training 4.3 Post-processing 5 Results 5.1 Segmentation 5.2 OS Prediction 6 Conclusion References Segmentation, Survival Prediction, and Uncertainty Estimation of Gliomas from Multimodal 3D MRI Using Selective Kernel Networks 1 Introduction 2 Methods 2.1 Preprocessing 2.2 Network Architecture 2.3 Loss Function 2.4 Optimization 2.5 Inference 2.6 Postprocessing 2.7 Overall Survival Prediction 2.8 Uncertainty Estimation 3 Results 3.1 Validation Set 3.2 Testing Set 4 Discussion and Conclusion References 3D Brain Tumor Segmentation and Survival Prediction Using Ensembles of Convolutional Neural Networks 1 Introduction 2 Method 2.1 Dataset 2.2 Preprocessing and Post-processing 2.3 Segmentation Pipeline 2.4 Survival Prediction Pipeline 2.5 Training 2.6 Uncertainty Estimation 3 Experiments and Results 3.1 Segmentation Performance 3.2 Survival Prediction 3.3 Evaluation of Uncertainty Measures in Segmentation 4 Discussion 5 Conclusions References Brain Tumour Segmentation Using Probabilistic U-Net 1 Introduction 2 Data 2.1 Preprocessing 3 Architecture 3.1 Probabilistic UNet 3.2 Attention 4 Training Details 5 Visualization and Analysis 5.1 Probabilistic UNet 5.2 Attention Maps 6 Results 7 Discussion and Future Scope 8 Conclusion References Segmenting Brain Tumors from MRI Using Cascaded 3D U-Nets 1 Introduction 2 Data 3 Methods 3.1 Data Standardization 3.2 Our U-Net-Based Architecture 3.3 Post-processing 3.4 Regularization Strategies 4 Experiments 4.1 Experimental Setup 4.2 Training Process 4.3 The Results 5 Conclusion References A Deep Supervised U-Attention Net for Pixel-Wise Brain Tumor Segmentation 1 Introduction 2 Method 2.1 Network Architecture 2.2 Evaluation Metrics 2.3 Loss Function 3 Experiments 3.1 Dataset Description 3.2 Data Pre-processing 3.3 Data Augmentation 3.4 Label Distribution 3.5 Training Procedure 4 Results 5 Conclusion References A Two-Stage Atrous Convolution Neural Network for Brain Tumor Segmentation and Survival Prediction 1 Introduction 2 Data 3 The Segmentation Algorithm 3.1 Brief Description of the Model 3.2 Details of the First Stage 3.3 Details of the Second Stage 3.4 Preprocessing 3.5 Training Details 3.6 Inference 3.7 Results 4 Overall Survival Prediction 4.1 Feature Extraction 4.2 Results 5 Conclusion References TwoPath U-Net for Automatic Brain Tumor Segmentation from Multimodal MRI Data 1 Introduction 2 Methods 2.1 Data Pre-processing and Augmentation 2.2 Network Architecture 2.3 Experiments 2.4 Network Training 3 Results 3.1 Evaluation Metrics 3.2 Preliminary Results 3.3 BraTS 2020 Challenge 4 Conclusion References Brain Tumor Segmentation and Survival Prediction Using Automatic Hard Mining in 3D CNN Architecture 1 Introduction 2 Related Work 3 Materials and Methods 3.1 Data 3.2 Task1: Brain Tumor Segmentation 3.3 Task2: Overall Survival Prediction 4 Results 4.1 Segmentation Task: 4.2 OS Prediction Task: 5 Conclusion References Some New Tricks for Deep Glioma Segmentation 1 Introduction 2 Related Work 3 Methods 3.1 Data Augmentation and Preprocessing 3.2 Architecture 3.3 Loss and Optimization 3.4 Model Ensembling for Uncertainty Estimation 4 Experimental Setup and Results References PieceNet: A Redundant UNet Ensemble 1 Introduction 1.1 Data 2 Methods 2.1 Modality Normalization 2.2 Label Map Separation 2.3 Patch Extraction 2.4 Network Structure and Training 2.5 Ensemble Construction 2.6 Volume Prediction 2.7 Survival Prediction 3 Results 3.1 Tumor Segmentation 3.2 Survival Prediction 4 Conclusion References Cerberus: A Multi-headed Network for Brain Tumor Segmentation 1 Introduction 2 Method 2.1 Cerberus Architecture 2.2 Loss Function 2.3 Pre-processing 2.4 Implementation Details 2.5 Inference 3 Experiments and Results 3.1 Database 3.2 Evaluation 3.3 Ablation Experiments 3.4 Comparison with the State-of-the-Art 3.5 Qualitative Results 4 Conclusions References An Automatic Overall Survival Time Prediction System for Glioma Brain Tumor Patients Based on Volumetric and Shape Features 1 Introduction 2 Materials 2.1 Dataset 2.2 Software and Hardware 3 Methods to Develop Overall Survival Time Prediction System 3.1 DL Segmentation Model 3.2 Features Extraction Methods 3.3 Overall Survival Time Prediction Model 3.4 Evaluation Methods 4 Experiments and Results 5 Conclusions and Future Work References Squeeze-and-Excitation Normalization for Brain Tumor Segmentation 1 Introduction 2 Materials and Methods 2.1 SE Normalization 2.2 Network Architecture 2.3 Data Preprocessing 2.4 Training Procedure 2.5 Loss Function 2.6 Ensembling 2.7 Post-processing 3 Results and Discussion References Modified MobileNet for Patient Survival Prediction 1 Introduction 2 Methodology 2.1 Dataset 2.2 Proposed Architecture 2.3 Preprocessing 2.4 Experiment Settings 2.5 Performance Evaluation 3 Result 4 Conclusion References Memory Efficient 3D U-Net with Reversible Mobile Inverted Bottlenecks for Brain Tumor Segmentation 1 Introduction 2 Methods 2.1 Reversible Layers 2.2 MobileNet Convolutional Block 2.3 Architecture 2.4 Training Procedure 3 Experiments and Results 4 Discussion References Brain Tumor Segmentation and Survival Prediction Using Patch Based Modified 3D U-Net 1 Introduction 2 Related Works 3 Methods 3.1 Datasets Description 3.2 Preprocessing 3.3 Patch Selection 3.4 Segmentation Pipeline and Network Structure 3.5 Overall Survival Prediction 4 Experimental Results and Discussion 4.1 Data and Implementation Details 4.2 Task 1: Brain Tumor Segmentation in MRI Scans 4.3 Task 2: Overall Survival (OS) Prediction from Pre-operative Scans 5 Conclusion References DR-Unet104 for Multimodal MRI Brain Tumor Segmentation 1 Introduction 2 Methods 2.1 Architecture 2.2 Loss Function 3 Experiment 3.1 Dataset, Pre-processing and Data Augmentation 3.2 Setup and Training 4 Results 5 Discussion 6 Conclusion References Glioma Sub-region Segmentation on Multi-parameter MRI with Label Dropout 1 Introduction 2 Related Work 3 Method 3.1 Preprocessing 3.2 Cropping and Data Augmentation 3.3 Model Architecture 3.4 Loss Function 3.5 Label Dropout 3.6 Implementation Details 3.7 Postprocessing 4 Results 5 Conclusion References Variational-Autoencoder Regularized 3D MultiResUNet for the BraTS 2020 Brain Tumor Segmentation 1 Introduction 2 Related Work 2.1 VAE-regularized 3D U-Net 2.2 MultiResUNet 3 Method 3.1 Encoder 3.2 Decoder with a VAE Branch 3.3 Res Path 3.4 Loss Function 3.5 Optimization 4 Experiment 4.1 Data Preprocessing and Augmentation 4.2 Results 5 Discussion and Conclusion References Learning Dynamic Convolutions for Multi-modal 3D MRI Brain Tumor Segmentation 1 Introduction 2 Methods 2.1 Multi-branch Dynamic Convolutional Networks 2.2 High-Level Features Learning Architecture 2.3 Loss Functions 2.4 Latest Further Works 3 Experiments 3.1 Datasets 3.2 Data Preprocessing 3.3 Implementation Details 3.4 Comparison with Previous Methods 3.5 Ablation Study 4 Conclusion References Computational Precision Medicine: Radiology-Pathology Challenge on Brain Tumor Classification Automatic Glioma Grading Based on Two-Stage Networks by Integrating Pathology and MRI Images 1 Introduction 2 Related Work 3 Method 3.1 Data Preprocessing 3.2 Model Details 4 Results and Discussion 4.1 Data and Evaluation Metrics 4.2 Experiments and Discussion 5 Conclusion References Brain Tumor Classification Based on MRI Images and Noise Reduced Pathology Images 1 Introduction 2 Related Work 3 Dataset and Method 3.1 Dataset 3.2 Preprocessing 3.3 Classification Based on MRI Images and Noise Reduced Pathology Images 4 Experiments 4.1 Implementation Details 4.2 Results and Discussion 5 Conclusion References Multimodal Brain Tumor Classification 1 Introduction 1.1 Related Work 2 Methods 2.1 Whole Slide Image Classification with a Generic and Modular Approach 2.2 Magnetic Resonance Imaging Classification 2.3 Multimodal Classification Through Ensembling 3 Experiments 3.1 Data 3.2 Implementation Details 3.3 Results 4 Conclusion and Discussion References A Hybrid Convolutional Neural Network Based-Method for Brain Tumor Classification Using mMRI and WSI 1 Introduction 2 Method 2.1 WSI-Based Method 2.2 mMRI-Based Method 3 Materials and Pre-processing 3.1 Data 3.2 Pre-processing for Radiology Images 4 Experiments and Results 4.1 Brain Tumor Segmentation in Radiology Images 4.2 Brain Tumor Classification Using Radiology and Pathology Images 4.3 Discussion 5 Conclusion References CNN-Based Fully Automatic Glioma Classification with Multi-modal Medical Images 1 Introduction 2 Related Works 3 Methods 3.1 Radiological Features Extraction 3.2 Pathological Features Extraction 3.3 Features Fusion Branch 3.4 Implementation and Training Details 4 Results 4.1 Quantitative Comparison 4.2 Timing Statistics 5 Conclusion References Glioma Classification Using Multimodal Radiology and Histology Data 1 Introduction 2 Related Work 3 Methodology 3.1 Dataset 3.2 Deep Learning Model: Densely Connected Network (DCN) 3.3 Histological Phase Analysis 3.4 Radiological Phase Analysis 3.5 Training DCNs 3.6 Outcome Integration and Final Sub-type Prediction 4 Results and Discussion 5 Conclusion References Author Index