[Communications in Computer and Information Science] Medical Image Understanding and Analysis Volume 1065 (23rd Conference, MIUA 2019, Liverpool, UK, July 24–26, 2019, Proceedings) ||

This book constitutes the refereed proceedings of the 23rd Conference on Medical Image Understanding and Analysis, MIUA 2019, held in Liverpool, UK, in July 2019. The 43 full papers presented were carefully reviewed and selected from 70 submissions. There were organized in topical sections named: oncology and tumour imaging; lesion, wound and ulcer analysis; biostatistics; fetal imaging; enhancement and reconstruction; diagnosis, classification and treatment; vessel and nerve analysis; image registration; image segmentation; ophthalmic imaging; and posters. Preface Organization Contents Oncology and Tumour Imaging Tissue Classification to Support Local Active Delineation of Brain Tumors 1 Introduction 2 Materials 2.1 Dataset 2.2 PET/CT Acquisition Protocol 3 Methods 3.1 The Proposed Segmentation 3.2 Sampling, Training and Performance of Classifiers 3.3 The Fully Automatic Protocol 3.4 The Modified Local Active Contour Method 3.5 Framework for Performance Evaluation 3.6 Gold Standard 4 Results 4.1 Classifier Validation 4.2 Clinical Testing and Results on Dataset 5 Discussion 6 Conclusion References Using a Conditional Generative Adversarial Network (cGAN) for Prostate Segmentation 1 Introduction 2 Methodology 2.1 Datasets 2.2 Model Architecture 2.3 Training and Post-processing 3 Results and Discussion 4 Conclusion 5 Future Work References A Novel Application of Multifractal Features for Detection of Microcalcifications in Digital Mammograms 1 Introduction 2 Materials and Methods 2.1 Microcalcifications Detection in Mammograms 2.2 Dataset 2.3 Processing Stages 3 Multifractal Analysis and Image Texture Enhancement 3.1 Multifractal Measures 3.2 Hölder Exponent and α-Image 3.3 Image Texture Enhancement 3.4 α-Value Range Selection 4 Microcalcifications Detection 4.1 Linear Structure Operator 4.2 MC Detection Based on Size 5 Experimental Results and Analysis 6 Conclusions References Wilms' Tumor in Childhood: Can Pattern Recognition Help for Classification? 1 Introduction 2 Materials and Methods 2.1 Data Sets 2.2 Features 3 Experiments 3.1 Nephroblastoma vs. Nephroblastomatosis 3.2 Subtype Determination 4 Conclusions References Multi-scale Tree-Based Topological Modelling and Classification of Micro-calcifications 1 Literature Review 2 Approach: Scale-Invariant Modelling of Micro-calcification 2.1 Dataset 2.2 Method 2.3 Results and Discussion 3 Comparison 4 Future Work 5 Conclusion References Lesion, Wound and Ulcer Analysis Automated Mobile Image Acquisition of Skin Wounds Using Real-Time Deep Neural Networks 1 Introduction 2 Related Work 3 System Overview 3.1 Image Focus Validation 3.2 Skin Wound Detection 3.3 Mobile Application 4 Results and Discussion 4.1 DNN Performance for Skin Wound Detection 4.2 User Interface Evaluation 5 Conclusions and Future Work References Hyperspectral Imaging Combined with Machine Learning Classifiers for Diabetic Leg Ulcer Assessment – A Case Study 1 Introduction 2 Materials and Methods 2.1 Patient 2.2 Hyperspectral Image Acquisition, Calibration and Processing 2.3 Hyperspectral Image Classification 3 Results 4 Discussion and Conclusion References Classification of Ten Skin Lesion Classes: Hierarchical KNN versus Deep Net 1 Introduction 2 Background 3 Edinburgh DERMOFIT Dataset 4 Hierarchical Classifier Methodology 4.1 Feature Calculation 4.2 Feature and Parameter Selection 4.3 Hierarchical Decision Tree 5 Decision Tree Experiment Results 6 Deep Net Classifier Methodology 7 Discussion References Biostatistics Multilevel Models of Age-Related Changes in Facial Shape in Adolescents 1 Introduction 2 Methods 2.1 Mathematical Formalism 2.2 Image Capture, Preprocessing, and Subject Characteristics 3 Results 4 Conclusions References Spatial Modelling of Retinal Thickness in Images from Patients with Diabetic Macular Oedema 1 Introduction 2 Methods 2.1 Image Dataset 2.2 Statistical Model 2.3 Spatial Dependency 2.4 Statistical Inference 2.5 Statistical Analysis 3 Results 4 Simulation 5 Discussion 6 Conclusion References Fetal Imaging Incremental Learning of Fetal Heart Anatomies Using Interpretable Saliency Maps 1 Introduction 2 Methodology 2.1 Saliency Map Quality 2.2 Saliency Driven Continual Curriculum 3 Model and Objective Functions 4 Results and Discussion References Improving Fetal Head Contour Detection by Object Localisation with Deep Learning 1 Introduction 2 Materials and Methods 2.1 Materials 2.2 Methods 3 Results and Discussion 4 Conclusions References Automated Fetal Brain Extraction from Clinical Ultrasound Volumes Using 3D Convolutional Neural Networks 1 Introduction 2 Brain Extraction Network 3 Experiments 3.1 Data 3.2 Performance Evaluation 4 Results 4.1 Cross-validation and Network Selection 4.2 Testing 4.3 Performance with Pose Variation 4.4 Performance with Different Gestational Ages 4.5 Regional Performance 4.6 Comparison with Previous Method 5 Discussion and Conclusion References Multi-task CNN for Structural Semantic Segmentation in 3D Fetal Brain Ultrasound 1 Introduction 2 Methods 2.1 Network Design 2.2 Label Generation 3 Results 4 Conclusion References Towards Capturing Sonographic Experience: Cognition-Inspired Ultrasound Video Saliency Prediction 1 Introduction 2 BDS-Net 3 Experiments 3.1 Data 3.2 Implementation Details 3.3 Evaluation Metrics 4 Results 4.1 Quantitative Results 4.2 Representative Examples 4.3 Ablation Study 5 Discussion 6 Conclusion and Outlook References Enhancement and Reconstruction A Novel Deep Learning Based OCTA De-striping Method 1 Introduction 2 Related Work 3 Method 3.1 Image Model 3.2 Model Architecture 3.3 Loss Function 4 Experiments and Discussion 4.1 Dataset 4.2 Results 4.3 Evaluation 4.4 Discussion 5 Conclusion References Edge Enhancement for Image Segmentation Using a RKHS Method 1 Introduction 2 RKHS 3 Proposed Segmentation Model Using RKHS 4 Numerical Experiments 5 Conclusion References A Neural Network Approach for Image Reconstruction from a Single X-Ray Projection 1 Introduction 2 Preliminaries 2.1 Clinical Feasibility 2.2 PCA Deformation Model 2.3 Neural Networks 3 Our Method 3.1 Training 3.2 Testing 4 Experiment 4.1 Simulated Data Sets 5 Conclusion References Deep Vectorization Convolutional Neural Networks for Denoising in Mammogram Using Enhanced Image 1 Introduction 2 Background 3 The Proposed Method 4 Experimental Results 4.1 Data 4.2 Experimental Results 5 Conclusion References Diagnosis, Classification and Treatment A Hybrid Machine Learning Approach Using LBP Descriptor and PCA for Age-Related Macular Degeneration Classification in OCTA Images 1 Motivation 2 Related Work 3 Proposed Approach 3.1 Texture Feature Extraction 3.2 Feature Dimensionality Reduction 3.3 Classification 4 Evaluation 4.1 Evaluation Setup and Criteria 4.2 Example Results 5 Conclusion and Future Work References Combining Fine- and Coarse-Grained Classifiers for Diabetic Retinopathy Detection 1 Introduction 1.1 Related Work 2 Materials and Methods 2.1 Datasets 2.2 Methodology 3 Results and Analysis 3.1 Results of Binary Classification 3.2 Results of Multi-class Classification 4 Conclusion References Vessel and Nerve Analysis Automated Quantification of Retinal Microvasculature from OCT Angiography Using Dictionary-Based Vessel Segmentation 1 Introduction 1.1 Related Work 2 Proposed Approach 2.1 Dictionary-Based Segmentation 2.2 Quantitative Analysis (Values VD, CD, BD, and VR) 3 Validation of Approach 3.1 Data and Scanning Protocol 3.2 Statistical Analysis 4 Results 5 Discussion 5.1 Discussion of Healthy Subjects 5.2 Discussion for Cataract Patients 6 Conclusion References Validating Segmentation of the Zebrafish Vasculature 1 Introduction 1.1 Zebrafish as a Model in Cardiovascular Research 1.2 Previous Work Aimed at Quantifying the Vascular System in Zebrafish 1.3 Contributions of This Work 2 Materials and Methods 2.1 Zebrafish Husbandry 2.2 Image Acquisition Settings, Properties and Data Analysis 2.3 Datasets 2.4 Image Enhancement, Segmentation and Total Vascular Volume Measurement 2.5 Tubular Filtering Enhancement Evaluation 2.6 Statistics and Data Representation 3 Results and Discussion 3.1 Sato Enhancement in Fiji 3.2 Influence of Sato Vessel Enhancement Scale on Measured Vessel Diameter 3.3 Assessment of Segmentation Robustness and Sensitivity 3.4 Conclusion References Analysis of Spatial Spectral Features of Dynamic Contrast-Enhanced Brain Magnetic Resonance Images for Studying Small Vessel Disease 1 Introduction 2 Methods 2.1 Analysis Framework 2.2 Subjects and Clinical Scores 2.3 Segmentation of Regions of Interest 2.4 Power Spectral Features of the Regions of Interest 2.5 Functional Data Analysis 2.6 Validation Against Clinical Parameters 3 Experiments and Results 4 Discussion References Segmentation of Arteriovenous Malformation Based on Weighted Breadth-First Search of Vascular Skeleton 1 Introduction 2 Methodology 2.1 Weighted Breadth-First Search Tree 2.2 Localization and Extraction of AVM Nidus 2.3 Segmentation of Feeding Arteries and Draining Veins 3 Experiments 3.1 Localization and Segmentation of AVM Nidus 3.2 Segmentation of Feeding Artery and Draining Vein 4 Conclusion References Image Registration A Variational Joint Segmentation and Registration Framework for Multimodal Images 1 Introduction 2 Related Work 3 The Proposed New Joint Regmentation Model for Multimodality 4 Numerical Results and Conclusions References An Unsupervised Deep Learning Method for Diffeomorphic Mono-and Multi-modal Image Registration 1 Introduction 2 A Learning Model 3 Numerical Tests 4 Conclusions References Automatic Detection and Visualisation of Metastatic Bone Disease 1 Introduction 2 Background 3 Methodology 3.1 Segmentation 3.2 Bone Mesh Creation and Cortical Bone Mapping (CBM) 3.3 Finding Symmetry 3.4 Atlas Creation 3.5 Articulated Registration 4 Visualisation of Disease 5 Results and Discussion 5.1 Evaluation 5.2 Discussion 6 Conclusions References High-Order Markov Random Field Based Image Registration for Pulmonary CT 1 Introduction 2 High-Order MRF Registration Model 2.1 General Form of the MRF Model 2.2 Potential Functions with Different Cliques 2.3 High-Order MRF Registration Model 3 Model Analysis 4 Results 5 Conclusions References Image Segmentation A Fully Automated Segmentation System of Positron Emission Tomography Studies 1 Introduction 2 Materials and Methods 2.1 Overview of the Proposed System 2.2 Phantom Studies 2.3 Clinical Studies 2.4 PET/CT Acquisition Protocol 2.5 Pre-processing of PET Dataset 2.6 Interesting Uptake Region Identification 2.7 The Enhanced Local Active Contour Method 2.8 Framework for Performance Evaluation 3 Results 3.1 Phantom Studies 3.2 Clinical Studies 4 Discussion References A General Framework for Localizing and Locally Segmenting Correlated Objects: A Case Study on Intervertebral Discs in Multi-modality MR Images 1 Introduction 2 Method 2.1 Step 1: Localizing and Labeling IVD Centroids 2.2 Step 2: Sampling Reoriented IVD Sections 2.3 Step 3: Segmenting IVD Tissue Using a V-Net 2.4 Step 4: Projecting Segmentations into Original Label Space 3 Results and Discussion 3.1 IVDM3Seg Data 3.2 Metrics 3.3 Evaluation on Released IVDM3Seg Data 3.4 Evaluation on Non-disclosed IVDM3Seg Data 4 Conclusions and Future Work References Polyp Segmentation with Fully Convolutional Deep Dilation Neural Network 1 Introduction 2 Related Work 3 Method 4 Implementation 4.1 Dataset 4.2 Data Augmentation 4.3 Evaluation Metrics 4.4 Cross-Validation Data 5 Results 5.1 Comparison with Benchmark Methods 5.2 Data Augmentation Ablation Tests 6 Conclusion References Ophthalmic Imaging Convolutional Attention on Images for Locating Macular Edema 1 Introduction 2 Visual Attention 2.1 Basic Visual Attention 2.2 Multi-class Attention 3 Experiments 3.1 Architectures 3.2 Results 4 Conclusion References Optic Disc and Fovea Localisation in Ultra-widefield Scanning Laser Ophthalmoscope Images Captured in Multiple Modalities 1 Introduction 2 Related Work 3 Materials 4 Method 4.1 Preprocessing 4.2 Architecture 4.3 Landmark Localisation Evaluation 4.4 Laterality Classification Evaluation 5 Results 6 Conclusion References Deploying Deep Learning into Practice: A Case Study on Fundus Segmentation 1 Introduction 2 Related Work 2.1 Generalization of ML Algorithms 2.2 Disc/Cup Segmentation 3 Methods 3.1 Databases 3.2 Pre- & Post-processing 3.3 Deep Learning 4 Results 5 Discussion and Conclusion References Joint Destriping and Segmentation of OCTA Images 1 Introduction 2 Methodology 2.1 Two-Stage Strategy for Segmenting OCTA Images Corrupted by Stripe Noise 2.2 A New Joint Model for Segmenting Images Corrupted by Stripe Noise 3 Results and Discussion 3.1 Effectiveness of Destriping 3.2 Effectiveness of Segmentation 4 Conclusion References Posters AI-Based Method for Detecting Retinal Haemorrhage in Eyes with Malarial Retinopathy 1 Introduction 2 Methodology 2.1 CNNs Architectures 2.2 Loss Functions 3 Experiments 3.1 Dataset 3.2 Data Preprocessing 3.3 Implementation 3.4 Evaluation Criteria 4 Results 5 Discussion 6 Conclusions References An Ensemble of Deep Neural Networks for Segmentation of Lung and Clavicle on Chest Radiographs 1 Introduction 2 Dataset 3 Method 3.1 Training U-Net 3.2 Training DeepLabv3+ Model 3.3 Fusing Segmentation Maps 3.4 Evaluation 4 Experiment and Results 5 Discussion 5.1 Ablation Study 5.2 Time Complexity 6 Conclusion References Automated Corneal Nerve Segmentation Using Weighted Local Phase Tensor 1 Introduction 1.1 Related Works 2 Method 2.1 Local Phase Tensor 2.2 Weighted Local Phase Tensor 3 Materials and Evaluation Metrics 3.1 Dataset 3.2 Evaluation Metrics 4 Experimental Results 5 Conclusions References Comparison of Interactions Between Control and Mutant Macrophages 1 Introduction 2 Materials 3 Methods 3.1 Movement Analysis Experiments 3.2 Selection of Experiments 4 Results 5 Discussion References Comparison of Multi-atlas Segmentation and U-Net Approaches for Automated 3D Liver Delineation in MRI 1 Introduction 2 Materials 3 Methods 3.1 Multi-atlas Segmentation Method 3.2 Deep Learning Segmentation Methods 3.3 Evaluation 4 Results 5 Discussion References DAISY Descriptors Combined with Deep Learning to Diagnose Retinal Disease from High Resolution 2D OCT Images 1 Introduction 2 Methods 2.1 Dataset 2.2 DAISY 2.3 Classification Network 2.4 Model Performance 3 Results 3.1 Training and Validation 3.2 Testing Dataset Performance 3.3 Comparisons 4 Discussion and Conclusions References Segmentation of Left Ventricle in 2D Echocardiography Using Deep Learning 1 Introduction 2 Methodology 2.1 Dataset 2.2 Neural Network for Semantic Segmentation 3 Evaluation Measures 4 Experiment Results and Discussion 5 Conclusion and Future Work References Author Index