Medical Image Computing and Computer Assisted Intervention – MICCAI 2022: 25th International Conference, Singapore, September 18–22, 2022, Proceedings, Part V (Lecture Notes in Computer Science)

✍ Scribed by Linwei Wang

• Tongue: English
• Leaves: 775
• Category: Library

✦ Table of Contents

Preface
Organization
Contents – Part V
Image Segmentation II
Automatic Segmentation of Hip Osteophytes in DXA Scans Using U-Nets
1 Introduction
2 Related Work
3 Method
4 Experiments
4.1 Data
4.2 Implementation Details
4.3 Results
5 Discussion and Conclusion
References
CIRDataset: A Large-Scale Dataset for Clinically-Interpretable Lung Nodule Radiomics and Malignancy Prediction
1 Introduction
2 CIRDataset
3 Method
3.1 Multi-class Mesh Decoder
3.2 Malignancy Prediction
4 Results and Discussion
References
UNeXt: MLP-Based Rapid Medical Image Segmentation Network
1 Introduction
2 UNeXt
3 Experiments and Results
4 Discussion
5 Conclusion
References
Exploring Smoothness and Class-Separation for Semi-supervised Medical Image Segmentation
1 Introduction
2 Method
2.1 Pixel-level Smoothness
2.2 Inter-class Separation
3 Experiment and Results
3.1 Dataset
3.2 Implementation Details
3.3 Results
4 Conclusion
References
Uncertainty-Guided Lung Nodule Segmentation with Feature-Aware Attention
1 Introduction
2 Methods
2.1 Uncertainty-Aware Module
2.2 Feature-Aware Attention Module
3 Experiments and Results
3.1 Dataset and Experimental Setup
3.2 Performance of General Segmentation
3.3 Performance of Multi-Confidence Mask
4 Conclusions
References
Thoracic Lymph Node Segmentation in CT Imaging via Lymph Node Station Stratification and Size Encoding
1 Introduction
2 Method
2.1 LN-Station Segmentation and Stratification
2.2 LN Size Stratification and Post-fusion
3 Experimental Results
4 Conclusion
References
ACT: Semi-supervised Domain-Adaptive Medical Image Segmentation with Asymmetric Co-training
1 Introduction
2 Methodology
2.1 Asymmetric Co-training for SSDA Segmentation
2.2 Pseudo-label with Exponential MixUp Decay
3 Experiments and Results
4 Conclusion
References
A Sense of Direction in Biomedical Neural Networks
1 Introduction
2 Related Works
3 Methods
3.1 Rotation Mechanism
3.2 Orientation and Scale Embedding
4 Experiments
5 Conclusion
References
Stroke Lesion Segmentation from Low-Quality and Few-Shot MRIs via Similarity-Weighted Self-ensembling Framework
1 Introduction
2 Method
2.1 Identify-to-Discern Network (IDN)
2.2 Soft Distribution-aware Updating (SDU)
3 Experiment
4 Conclusion
References
Edge-Oriented Point-Cloud Transformer for 3D Intracranial Aneurysm Segmentation
1 Introduction
2 Approach
2.1 Dual Stream Transformer (DST)
2.2 Outer-edge Context Dissimilation (OCD)
2.3 Inner-Edge Hard-Sample Excavation (IHE)
2.4 Training and Testing
3 Experiment
4 Conclusions
References
mmFormer: Multimodal Medical Transformer for Incomplete Multimodal Learning of Brain Tumor Segmentation
1 Introduction
2 Method
2.1 Hybrid Modality-Specific Encoder
2.2 Modality-Correlated Encoder
2.3 Convolutional Decoder
2.4 Auxiliary Regularizer
3 Experiments and Results
4 Conclusion
References
Multimodal Brain Tumor Segmentation Using Contrastive Learning Based Feature Comparison with Monomodal Normal Brain Images
1 Introduction
2 Method
2.1 The Normal Appearance Network
2.2 The Segmentation Backbone
3 Experiments
3.1 Evaluation of Segmentation Results
3.2 Evaluation of Contrastive Learning Based Feature Comparison
4 Conclusion
References
Meta-hallucinator: Towards Few-Shot Cross-Modality Cardiac Image Segmentation
1 Introduction
2 Method
2.1 Gradient-Based Meta-hallucination Learning
2.2 Hallucination-Consistent Self-ensembling Learning
3 Experiments and Results
4 Conclusions
References
NestedFormer: Nested Modality-Aware Transformer for Brain Tumor Segmentation
1 Introduction
2 Method
2.1 Global Poolformer Encoder
2.2 Nested Modality-Aware Feature Aggregation
2.3 Modality-Sensitive Gating
3 Experiment
3.1 Implementation Details
3.2 Datasets and Evaluation Metrics
3.3 Comparison with SOTA Methods
3.4 Ablation Study
4 Conclusion
References
MaxStyle: Adversarial Style Composition for Robust Medical Image Segmentation
1 Introduction
2 Related Work
3 Method
3.1 Preliminaries: MixStyle
3.2 Robust Feature Learning and Improved Interpretability with Auxiliary Image Decoder
3.3 MaxStyle
4 Experiments and Results
4.1 Data: Cardiac MR Segmentation Datasets
4.2 Implementation and Experiment Set-Up
4.3 Results
5 Discussion and Conclusion
References
A Robust Volumetric Transformer for Accurate 3D Tumor Segmentation
1 Introduction
2 Methodology
3 Related Work
4 Experiments
5 Conclusion
References
Usable Region Estimate for Assessing Practical Usability of Medical Image Segmentation Models
1 Introduction
2 Methodology
2.1 Preliminaries
2.2 Correctness-Confidence Rank Correlation (CCRC)
2.3 Usable Region Estimate (URE)
3 Experiments
3.1 CCRC and URE Provide New Insights for Model Evaluation
3.2 Estimated Usable Regions on New Unseen Samples
4 Conclusions
References
Modality-Adaptive Feature Interaction for Brain Tumor Segmentation with Missing Modalities
1 Introduction
2 Method
2.1 Modality-Adaptive Feature Interaction
2.2 Network Details and Training Loss
3 Experiments
4 Conclusion
References
Position-Prior Clustering-Based Self-attention Module for Knee Cartilage Segmentation
1 Introduction
2 Method
3 Experiment
4 Conclusion
References
Attentive Symmetric Autoencoder for Brain MRI Segmentation
1 Introduction
2 Methodology
2.1 Attentive Symmetric Autoencoder
2.2 Network Architecture
3 Experiments and Results
4 Conclusion
References
Denoising for Relaxing: Unsupervised Domain Adaptive Fundus Image Segmentation Without Source Data
1 Introduction
2 Methods
2.1 Pseudo Labeling with Class-Dependent Thresholds
2.2 Label Self-correction Towards Effective Adaptation
3 Experiments
3.1 Datasets
3.2 Implementation Details and Evaluation Metrics
3.3 Results
4 Conclusion
References
Curvature-Enhanced Implicit Function Network for High-quality Tooth Model Generation from CBCT Images
1 Introduction
2 Method
2.1 High-quality Tooth Model Building
2.2 Tooth Instance Segmentation
2.3 Surface Reconstruction
2.4 Curvature Enhancement
3 Experiments
3.1 Dataset and Evaluation Metrics
3.2 Sampling Strategies
3.3 Implementation Details
3.4 Comparison with Other Methods
3.5 Ablation Study
4 Conclusion
References
PHTrans: Parallelly Aggregating Global and Local Representations for Medical Image Segmentation
1 Introduction
2 Method
2.1 Overall Architecture
2.2 Parallel Hybrid Module
3 Experiments
3.1 Dataset
3.2 Implementation Details
3.3 Results
4 Conclusions
References
Learning Tumor-Induced Deformations to Improve Tumor-Bearing Brain MR Segmentation
1 Introduction
2 Method
2.1 Deformation Network
2.2 Synthesizing Tumor Images for Training
2.3 Modifying the Atlas
3 Experiments
3.1 Inter-Hemisphere Symmetry Validation
3.2 Synthesized Images Evaluation
3.3 Test with Real Data
4 Conclusion
References
Contrastive Re-localization and History Distillation in Federated CMR Segmentation
1 Introduction
2 Methodology
2.1 CRL for Representation Bias Correction:
2.2 MD for Continuous Optimization of Each Center
3 Experiments and Results
4 Discussion and Conclusion
References
Contrast-Free Liver Tumor Detection Using Ternary Knowledge Transferred Teacher-Student Deep Reinforcement Learning
1 Introduction
2 Methodology
2.1 Network Architecture
2.2 Ternary Knowledge Set
2.3 P-strategy
3 Experiment
3.1 Experimental Setup
3.2 Experimental Results
4 Conclusion
References
DeepPyramid: Enabling Pyramid View and Deformable Pyramid Reception for Semantic Segmentation in Cataract Surgery Videos
1 Introduction
2 Methodology
3 Experimental Setup
4 Experimental Results
5 Conclusion
References
A Geometry-Constrained Deformable Attention Network for Aortic Segmentation
1 Introduction
2 Geometry-Constrained Deformable Attention Network
2.1 Problem Formulation
2.2 Deformable Attention Extractor
2.3 Geometry-Constrained Guider
3 Experiment and Result
4 Conclusion
References
ConTrans: Improving Transformer with Convolutional Attention for Medical Image Segmentation
1 Introduction
2 Method
2.1 Swin Transformer Encoder
2.2 Depthwise Attention Block
2.3 Spatial-Reduction-Cross-Attention
3 Experiments
3.1 Datasets
3.2 Implementation Details
3.3 Comparisons with State-of-the-Art Methods
3.4 Ablation Studies
4 Conclusions
References
End-to-End Segmentation of Medical Images via Patch-Wise Polygons Prediction
1 Introduction
2 Related Work
3 Methods
4 Experiments
5 Conclusions
References
Automatic Identification of Segmentation Errors for Radiotherapy Using Geometric Learning
1 Introduction
2 Materials and Method
2.1 Dataset
2.2 Generating a Training Dataset
2.3 A Hybrid CNN-GNN Model for Contour Error Prediction
2.4 CNN Pre-training
2.5 Implementation Details
3 Experiments
3.1 Ablation Tests
4 Results
4.1 Ablation Tests
5 Discussion
6 Conclusion
References
A Novel Knowledge Keeper Network for 7T-Free but 7T-Guided Brain Tissue Segmentation
1 Introduction
2 Proposed Method
2.1 Teacher Network
2.2 Knowledge Keeper Network
2.3 Brain Tissue Segmentation Network
3 Experimental Settings and Results
3.1 Dataset
3.2 Results and Analysis
4 Conclusion
References
OnlyCaps-Net, a Capsule only Based Neural Network for 2D and 3D Semantic Segmentation
1 Introduction
2 Method
3 Experimental Setup and Results
4 Discussion and Conclusion
References
Identifying and Combating Bias in Segmentation Networks by Leveraging Multiple Resolutions
1 Introduction
2 Methodology
2.1 Networks
3 Experiments
3.1 Task 1 – Cortical Segmentation in Adults and Children
3.2 Task 2 – Stratified Hippocampal Segmentation
3.3 Evaluation Measures
3.4 Training
4 Results
5 Conclusion
References
Joint Modeling of Image and Label Statistics for Enhancing Model Generalizability of Medical Image Segmentation
1 Introduction
2 Methodology
2.1 Statistical Modeling of Image and Label
2.2 Variational Inference of Image and Label
2.3 Neural Networks and Training Strategy
3 Experiments
3.1 Tasks and Datasets
3.2 Cross-Sequence Segmentation
3.3 Cross-Site Segmentation
3.4 Interpretation of Joint Modeling
4 Conclusion
References
Transformer Based Feature Fusion for Left Ventricle Segmentation in 4D Flow MRI
1 Introduction
2 Method
2.1 Attention Mechanism
2.2 Feature Fusion Layer
2.3 Network Structure
3 Materials
3.1 Dataset
3.2 Evaluation Metrics
4 Experiment and Results
5 Conclusion
References
Learning Towards Synchronous Network Memorizability and Generalizability for Continual Segmentation Across Multiple Sites
1 Introduction
2 Method
2.1 Synchronous Gradient Alignment (SGA)
2.2 Efficient Optimization by Dual-Meta Algorithm
2.3 Comprehensive Configuration of Replay Buffer
3 Experiments
4 Conclusion
References
Progressive Deep Segmentation of Coronary Artery via Hierarchical Topology Learning
1 Introduction
2 Method
2.1 Spatial Anatomical Dependency
2.2 Hierarchical Topology Learning
2.3 Loss Function
3 Experiments and Results
3.1 Dataset and Evaluation Metrics
3.2 Implementation Details
3.3 Comparisons with the State-of-the-Art Methods
3.4 Ablation Study
4 Conclusion
References
Evidence Fusion with Contextual Discounting for Multi-modality Medical Image Segmentation
1 Introduction
2 Methods
2.1 Evidential Segmentation
2.2 Multi-modality Evidence Fusion
2.3 Discounted Dice Loss
3 Experimental Results
3.1 Experiment Settings
3.2 Segmentation Results
4 Conclusion
References
Orientation-Guided Graph Convolutional Network for Bone Surface Segmentation
1 Introduction
2 Method
2.1 Graph Convolution for Bone Segmentation
2.2 Bone Orientation Learning as an Auxiliary Task
2.3 Network Details
3 Experiments and Results
3.1 Dataset
3.2 Implementation Details
3.3 Evaluation Metrics
3.4 Quantitative Comparison
3.5 Qualitative Results
3.6 Ablation Study
4 Discussion
4.1 Bone Orientation Loss
4.2 CNN Vs. GCN
5 Conclusion
References
Weakly Supervised Volumetric Image Segmentation with Deformed Templates
1 Introduction
2 Related Work
2.1 Weakly-Supervised Image Segmentation
2.2 Template Based Approaches
2.3 Image Reconstruction
3 Method
3.1 Network Architecture and Losses
3.2 Template Deformation
4 Experiments
4.1 Datasets, Metrics and Baseline
4.2 Comparative Results
4.3 Human Annotations
5 Conclusion
References
Implicit Neural Representations for Medical Imaging Segmentation
1 Introduction
2 Proposed Approach
2.1 Method Overview
2.2 Implementation
3 Experiments
3.1 Segmentation Performance and Memory Requirements
3.2 Small Organs and Model Convergence
3.3 Super-Resolution
3.4 Dataset Splits and Inference Efficiency
3.5 Network Architectures and Sampling Strategies
4 Conclusion
References
ModDrop++: A Dynamic Filter Network with Intra-subject Co-training for Multiple Sclerosis Lesion Segmentation with Missing Modalities
1 Introduction
2 Methods
2.1 Preliminaries
2.2 Dynamic Head with Filter Scaling
2.3 Intra-subject Co-training
2.4 Implementation Details
2.5 Datasets and Evaluation Metrics
3 Experimental Results
4 Discussion and Conclusion
References
DOMINO: Domain-Aware Model Calibration in Medical Image Segmentation
1 Introduction
2 Domain-Aware Model Calibration
2.1 U-Net Transformers (UNETR) Model
2.2 Domain-Aware Loss Regularization
3 Experiments and Results
3.1 Dataset
3.2 Evaluation Metrics
3.3 Calibrated Models Outperform UNETR-Base on 11-Classes
3.4 Calibrated UNETR Outperforms or Performs Comparably to Headreco in 6-Class Segmentation
4 Conclusions
References
iSegFormer: Interactive Segmentation via Transformers with Application to 3D Knee MR Images
1 Introduction
2 Related Work
3 Method
3.1 Network Architecture of iSegFormer
3.2 Clicks Encoding and Simulation
3.3 Training and Inference Details
3.4 Extending to Interactive 3D Image Segmentation
4 Experiments
4.1 Results of Interactive 2D Image Segmentation
4.2 Results on Segmentation Propagation
4.3 Ablation Study
5 Conclusion
References
Patcher: Patch Transformers with Mixture of Experts for Precise Medical Image Segmentation
1 Introduction
2 Method
2.1 Overall Architecture
2.2 Patcher Block
2.3 Patcher Encoder
2.4 Mixture of Experts Decoder
3 Experiments
3.1 Comparison with State-of-the-Art Methods
3.2 Ablation Study
4 Conclusions
References
TransFusion: Multi-view Divergent Fusion for Medical Image Segmentation with Transformers
1 Introduction
2 Methods
2.1 Revisiting the Self-attention Mechanism
2.2 Divergent Fusion Attention Module (DiFA)
2.3 Multi-Scale Attention (MSA) Module
3 Experiments
3.1 Datasets and Settings
3.2 Results
3.3 Ablation Studies
4 Conclusion
References
CorticalFlow++: Boosting Cortical Surface Reconstruction Accuracy, Regularity, and Interoperability
1 Introduction
2 Related Work
3 Method
3.1 CorticalFlow Framework
3.2 Higher Order ODE Solver
3.3 Smooth Templates
3.4 White to Pial Surface Morphing
4 Experiments
5 Conclusion
References
Carbon Footprint of Selecting and Training Deep Learning Models for Medical Image Analysis
1 Introduction
2 Related Work
3 Methods for Carbon Footprinting
4 Data and Experiments
5 Discussions
6 Conclusions
References
SMESwin Unet: Merging CNN and Transformer for Medical Image Segmentation
1 Introduction
2 SMESwin Unet for Medical Image Segmentation
2.1 Encoder-Decoder Architecture
2.2 Superpixel
2.3 MCCT
2.4 External Attention
3 Experiments and Results
3.1 Dataset
3.2 Implementation Details
3.3 Experiments
4 Conclusion
References
The Dice Loss in the Context of Missing or Empty Labels: Introducing and
1 Introduction
2 Bells and Whistles of the Dice Loss: and
2.1 Configuration of and in Practice
2.2 Effect of and on Missing or Empty Labels
2.3 A Simple Heuristic for Tuning to Learn from Empty Maps
3 Experimental Setup
4 Results
5 Discussion
6 Conclusion
References
Robust Segmentation of Brain MRI in the Wild with Hierarchical CNNs and No Retraining
1 Introduction
2 Methods
2.1 Hierarchical Conditional Architecture
2.2 Training Scheme for the Segmentation Modules
2.3 Training Scheme for the Denoising Module
2.4 Implementation Details
3 Experiments and Results
3.1 Brain MRI Datasets
3.2 Competing Methods
3.3 Quantitative Analysis
3.4 Volumetric Study
4 Conclusion
References
Deep Reinforcement Learning for Small Bowel Path Tracking Using Different Types of Annotations
1 Introduction
2 Method
2.1 Dataset
2.2 Environment
2.3 Training and Testing
2.4 Evaluation Details
3 Results
3.1 Quantitative Evaluation
3.2 Qualitative Evaluation
4 Conclusion
References
Efficient Population Based Hyperparameter Scheduling for Medical Image Segmentation
1 Introduction
2 Method
3 Experiments
4 Discussion and Conclusion
References
Atlas-Based Semantic Segmentation of Prostate Zones
1 Introduction
2 Methods
2.1 Prostate Zone Atlas Construction and Image Pre-processing
2.2 Atlas-Based Semantic Segmentation Architecture
2.3 Model Training
2.4 Model Inference
3 Experiments and Results
4 Discussion and Conclusion
References
Bayesian Pseudo Labels: Expectation Maximization for Robust and Efficient Semi-supervised Segmentation
1 Introduction
2 Pseudo Labelling as Expectation-Maximization
3 Generalisation of Pseudo Labels via Variational Inference for Segmentation
4 Experiments and Results
4.1 SegPL Outperforms Baselines with Less Computational Resource and Less Training Time
4.2 Ablation Studies
4.3 Better Generalisation on Out-of-Distribution (OOD) Samples and Better Robustness Against Adversarial Attack
4.4 Uncertainty Estimation with SegPL-VI
5 Related Work
6 Conclusion
References
Enforcing Connectivity of 3D Linear Structures Using Their 2D Projections
1 Introduction
2 Related Work
3 Approach
3.1 Connectivity Loss
3.2 Projected Connectively Loss
3.3 Total Loss
4 Experiments
4.1 Datasets
4.2 Metrics
4.3 Architectures and Baselines
4.4 Results
5 Conclusion and Future Work
References
RT-DNAS: Real-Time Constrained Differentiable Neural Architecture Search for 3D Cardiac Cine MRI Segmentation
1 Introduction
2 Method
2.1 Search Space and Method
2.2 Latency Constraint Incorporation
2.3 Architecture Derivation
3 Experiments
3.1 Experiment Setup
3.2 Results
4 Conclusion
References
Integration of Imaging with Non-imaging Biomarkers
Lesion Guided Explainable Few Weak-Shot Medical Report Generation
1 Introduction
2 Methods
3 Experiments
4 Conclusion
References
Survival Prediction of Brain Cancer with Incomplete Radiology, Pathology, Genomic, and Demographic Data
1 Introduction
2 Methods
2.1 Uni-Modal Feature Embedding
2.2 Multi-modal Fusion with Complete Data
2.3 Multi-modal Fusion for Missing Data
3 Data and Experimental Setting
4 Results
5 Ablation Studies for Answering Four Questions
6 Conclusion
References
Discrepancy and Gradient-Guided Multi-modal Knowledge Distillation for Pathological Glioma Grading
1 Introduction
2 Method
2.1 Discrepancy-Induced Contrastive Distillation
2.2 Gradient-Guided Knowledge Refinement
3 Experiments
3.1 Dataset and Implementation Details
3.2 Experimental Results
4 Conclusion
References
Radiological Reports Improve Pre-training for Localized Imaging Tasks on Chest X-Rays
1 Introduction and Motivation
2 Material and Methods
2.1 Studied Pre-training Methods
2.2 Evaluation Framework
3 Results
4 Discussion
5 Conclusion
References
Anatomy-Guided Weakly-Supervised Abnormality Localization in Chest X-rays
1 Introduction
2 Methods
3 Experiments
4 Conclusion
References
Identification of Vascular Cognitive Impairment in Adult Moyamoya Disease via Integrated Graph Convolutional Network
1 Introduction
2 Method
2.1 Materials and Preprocessing
2.2 Graph Construction with Different Modalities
2.3 Dual-Modal GCN Module
2.4 Node-Based Normalization and Constrain
3 Experiments and Results
3.1 Experimental Settings
3.2 Identification Results
3.3 Biomarker Interpretation
4 Conclusion
References
Multi-modal Masked Autoencoders for Medical Vision-and-Language Pre-training
1 Introduction
2 The Proposed Approach
2.1 The Backbone Model Architecture
2.2 Multi-modal Masked Autoencoders
3 Experiments
3.1 Pre-training Setup
3.2 Vision-and-Language Transfer Tasks
3.3 Comparisons with the State-of-the-Art
3.4 Quantitative Analysis
3.5 Qualitative Analysis
4 Conclusion
References
Breaking with Fixed Set Pathology Recognition Through Report-Guided Contrastive Training
1 Introduction
2 Global-Local Contrastive Learning
2.1 Model Overview
2.2 Training Objectives
2.3 Model Inference
3 Prompt Engineering
4 Experiments
4.1 Experimental Setup
4.2 Results
5 Conclusion
References
Explaining Chest X-Ray Pathologies in Natural Language
1 Introduction
2 MIMIC-NLE
3 Models
4 Experimental Setup
5 Results and Discussion
6 Summary and Outlook
A Supplementary
References
RepsNet: Combining Vision with Language for Automated Medical Reports
1 Introduction
2 Related Work
3 RepsNet: Proposed Approach
3.1 Contrastive Image-Text Encoder
3.2 Conditional Language Decoder
4 Experiments, Results and Discussion
4.1 Visual Question Answering
4.2 Medical Report Generation
5 Conclusion
References
BERTHop: An Effective Vision-and-Language Model for Chest X-ray Disease Diagnosis
1 Introduction
2 Related Work
3 Approach
3.1 Visual Encoder
3.2 In-Domain Text Pre-training
4 Experiments
4.1 Experiment Setup
4.2 Main Results
4.3 In-Domain Text Pre-training
4.4 Visual Encoder
5 Discussion and Conclusion
References
Author Index