Multimodality Imaging, Volume 1 : Deep Learning Applications

This research and reference text explores the finer details of deep learning models. It provides a brief outline on popular models including convolution neural networks, deep belief networks, autoencoders and residual neural networks. The text discusses some of the deep learning-based applications in gene identification. Sections in the book explore the foundation and necessity of deep learning in radiology, application of deep learning in the area of cardiovascular imaging and deep learning applications in the area of fatty liver disease characterization and COVID-19, respectively. This reference text is highly relevant for medical professionals and researchers in the area of artificial intelligence in medical imaging. Key features • Discusses various diseases related to lung, heart, peripheral arterial imaging, as well as gene expression characterization and classification. • Explores imaging applications, their complexities and the deep learning models employed to resolve them in detail. • Provides state-of-the-art contributions while addressing doubts in multimodal research. • Details the future of deep learning and big data in medical imaging. PRELIMS.pdf Preface Purpose Content and organization Editor biographies Mainak Biswas Jasjit S Suri List of contributors CH001.pdf Chapter 1 Deep learning and augmented radiology 1.1 Introduction 1.2 The present 1.2.1 Fatty liver disease risk stratification 1.2.2 Carotid intima–media thickness (cIMT) measurement using DL for segmentation 1.2.3 Assessment of the treatment effects in acute ischemic stroke (AIS) using DL from MR images 1.2.4 Diagnosis of prostate cancer using DL 1.2.5 CT-based respiratory disease prognosis using DL 1.3 Future 1.3.1 DL in radiology 1.3.2 The future of radiology 1.4 The potential of deep learning 1.4.1 Economy 1.4.2 Augmented radiology and DL 1.4.3 Further development of DL 1.5 Challenges and risks in DL 1.5.1 Safety 1.5.2 Privacy 1.5.3 Legality 1.6 Conclusion References CH002.pdf Chapter 2 Deep learning in biomedical imaging 2.1 Introduction 2.2 Deep learning models 2.2.1 Deep belief networks 2.2.2 Autoencoder 2.2.3 Convolutional neural networks 2.2.4 Deep residual network 2.3 DL based biomedical imaging systems 2.3.1 Cardiovascular application 2.3.2 Neurology application 2.3.3 Mammography application 2.3.4 Microscopy applications 2.3.5 Dermatology application 2.3.6 Gastroenterology applications 2.3.7 Pulmonary application 2.4 Discussion 2.4.1 Graphics processing unit and open-source software for deep learning References CH003.pdf Chapter 3 A review of artificial intelligence in brain tumor classification and segmentation 3.1 Introduction 3.2 Brain cancer pathophysiology 3.2.1 Architecture at the cellular level 3.2.2 Links between brain tumors and genes 3.3 Imaging modality 3.3.1 Computed tomography imaging 3.3.2 Magnetic resonance imaging 3.3.3 Biopsy 3.3.4 Hyperstereoscopy imaging 3.3.5 MR spectroscopy 3.4 Guidelines for tumor grading by the WHO 3.5 Brain tumor tests 3.5.1 Biomarker test 3.5.2 Biopsy 3.5.3 Imaging test 3.6 Classification methods 3.6.1 Machine learning 3.6.2 Deep learning 3.6.3 Brain image analysis using deep learning 3.6.4 A plausible solution for brain cancer classification 3.7 Brain cancer and other brain disorders 3.7.1 Stroke 3.7.2 Alzheimer’s disease 3.7.3 Parkinson’s disease 3.7.4 Leukoaraiosis 3.7.5 Multiple sclerosis 3.7.6 Wilson’s disease 3.8 Discussion 3.8.1 A note on cancer detection biomarkers 3.9 Conclusion References CH004.pdf Chapter 4 MRI based brain tumor classification and its validation: a transfer learning paradigm 4.1 Introduction 4.2 Background literature survey 4.3 Demographics and data preparation 4.3.1 Patient demographics 4.3.2 Data preparation 4.4 Methodology 4.4.1 CNN model 4.4.2 The architecture of AlexNet 4.4.3 Transfer learning and workflow 4.4.4 Weight optimization 4.4.5 A generalization system for tumor classification 4.5 Experimental protocol, results, and performance evaluation 4.5.1 Parameter selection and simulation 4.5.2 Results 4.5.3 Performance evaluation 4.6 Model validation and verification 4.6.1 Hypothesis validation 4.6.2 Software verification 4.7 Discussion 4.8 Conclusion Appendix A References CH005.pdf Chapter 5 Magnetic resonance based Wilson’s disease tissue characterization in an artificial intelligence framework using transfer learning 5.1 Introduction 5.2 Background literature 5.3 Methodology 5.3.1 Patient demographics 5.3.2 Data augmentation 5.3.3 Pre-processing: skull and background removal 5.4 Global architecture: transfer learning 5.4.1 AlexNet 5.4.2 ResNet50 5.4.3 DenseNet161 5.4.4 XceptionNet 5.4.5 InceptionV3 5.4.6 SivaSuriNet 5.5 Results 5.6 Characterization 5.6.1 Mean feature strength 5.6.2 Higher-order spectrum 5.7 Discussion 5.8 Conclusion References CH006.pdf Chapter 6 Artificial intelligence based carotid plaque tissue characterisation and classification from ultrasound images using a deep learning paradigm 6.1 Introduction 6.2 Methodology 6.2.1 Patient demographics 6.2.2 Exclusion criteria 6.2.3 Ultrasound data acquisition and preprocessing 6.2.4 Plaque delineation 6.2.5 Ultrasound plaque data augmentation 6.2.6 Supercomputer specifications 6.2.7 Deep learning architecture 6.2.8 Experimental protocol 6.2.9 Machine learning for benchmarking deep learning 6.2.10 Performance parameters using the DL and ML methods 6.3 Results 6.3.1 Deep learning data analysis and benchmarking against machine learning 6.3.2 Plaque characterisation in a deep learning framework 6.4 Discussion 6.4.1 A note on the unbalanced datasets for symptomatic and asymptomatic plaques 6.4.2 Benchmarking against techniques available in the literature 6.4.3 A special note on the comparison of supercomputer hardware to a local machine 6.4.4 Strengths, weaknesses, and extensions 6.5 Conclusion Disclosure/Conflict of interest Appendix References CH007.pdf Chapter 7 Quantification of plaque volume using a two-stage deep learning paradigm 7.1 Introduction 7.2 Background 7.3 Data acquisition 7.4 Methodology 7.5 Experimental protocol and results 7.6 Statistical tests 7.7 Discussion 7.8 Conclusion References CH008.pdf Chapter 8 Stenosis measurement from ultrasound carotid artery images in the deep learning paradigm 8.1 Introduction 8.2 Patient demographics and image acquisition 8.3 Methodology 8.4 Experimental protocol, performance parameters, and results 8.5 Statistical tests, variability and error bias analysis, and risk characterization 8.6 Discussion 8.7 Conclusion References CH009.pdf Chapter 9 A systematic review of conventional and deep learning models for the measurement of plaque burden 9.1 Introduction 9.2 Chronological generation of cIMT regional segmentation and cIMT measurement 9.3 Ml application for cIMT and PA measurement 9.3.1 ANN model for cIMT region detection 9.3.2 Extreme learning machine radial basis neural network model for cIMT region detection 9.3.3 Fuzzy K-means classifier for cIMT region extraction 9.4 Deep learning application for cIMT and PA extraction 9.4.1 ANN autoencoder based cIMT region segmentation 9.4.2 Fully convolutional network for cIMT region estimation 9.4.3 FCN for PA measurement 9.4.4 Two-stage patching based AI model for cIMT and PA measurement 9.5 Discussion 9.5.1 Benchmarking 9.5.2 A short note on cardiovascular risk assessment 9.5.3 A note on the clinical impact of AI methods on cIMT/PA techniques 9.5.4 A note on inter- and intra-observer variability analysis on the evaluation of AI models 9.5.5 A short note on 10 year risk estimation using cIMT and PA 9.5.6 Statistical power analysis and diagnostic-odds ratio 9.6 Conclusions Appendix A Mathematical representations of the ML and DL paradigms References CH010.pdf Chapter 10 Ultrasound fatty liver disease risk stratification using an extreme learning machine framework 10.1 Introduction 10.2 Data demographics, collection, and preparation 10.2.1 Sub-sampling of US datasets (S4 and S8) 10.3 Methodology 10.3.1 The three-layered ELM architecture 10.3.2 Tissue characterization and risk stratification using ELM and SVM frameworks 10.3.3 Feature mining 10.4 Experimental protocol 10.4.1 Experiment 1: the effect of the size of the training data on accuracy using four CV protocols 10.4.2 Experiment 2: the effect of the training set size using the sub-sampling strategy 10.4.3 Experiment 3: ELM and SVM time complexity 10.5 Results 10.5.1 Experiment 1: the effect of training data size on accuracy using the four CV protocols 10.5.2 Experiment 2: the effect of the training data size in parts during the CV protocols 10.5.3 Experiment 3: time comparison between ELM and SVM 10.6 Performance evaluations 10.6.1 ROC curves 10.6.2 Reliability and stability analysis 10.7 Discussion 10.7.1 Benchmarking 10.7.2 ELM and BPNN comparison 10.7.3 A special note on the ELM and SVM 10.7.4 Strengths, weaknesses, and future work 10.8 Conclusions Appendix A Scientific validation Appendix B Results of the ELM/SVM classifier for the S4 and S8 datasets References CH011.pdf Chapter 11 Symtosis: deep learning based liver ultrasound tissue characterisation and risk stratification 11.1 Introduction 11.2 Patient demographics and acquisition 11.3 Methodology 11.3.1 Risk stratification model 11.3.2 CNN architecture 11.4 Results 11.4.1 Image pre-processing for the DL, ELM, and SVM 11.4.2 The effect of data size on stratification accuracy 11.4.3 Stratification analysis using the ‘liver segregation index’ 11.5 Performance evaluation: ROC, reliability, and timing analysis 11.5.1 ROC analysis 11.5.2 Reliability analysis 11.5.3 Timing analysis 11.6 Discussion 11.7 Conclusion References CH012.pdf Chapter 12 Characterization of COVID-19 severity in infected lungs via artificial intelligence transfer learning 12.1 Introduction 12.2 Methodology 12.2.1 Patient demographics 12.2.2 Data acquisition 12.2.3 Baseline characteristics 12.2.4 Segmentation 12.2.5 Augmentation 12.2.6 Models 12.3 Results 12.4 Characterization 12.5 Discussion References This research and reference text explores the finer details of Deep Learning models. It provides a brief outline on popular models including convolution neural networks (CNN), deep belief networks (DBN), autoencoders, residual neural networks (Res Nets). The text discusses some of the Deep Learning-based applications in gene identification. Sections in the book explore the foundation and necessity of deep learning in radiology, the application of deep learning in the area of cardiovascular imaging and deep learning applications in the area of fatty liver disease characterization and COVID19, respectively. This reference text is highly relevant for medical professionals and researchers in the area of AI in medical imaging. Key Features:Discusses various diseases related to lung, heart, peripheral arterial imaging, as well as gene expression characterization and classificationExplores imaging applications, their complexities and the Deep Learning models employed to resolve them in detailProvides state-of-the-art contributions while addressing doubts in multimodal researchDetails the future of deep learning and big data in medical imaging