State of the Art in Neural Networks and Their Applications: Volume 2

Front Cover
State of the Art in Neural Networks and Their Applications
Copyright Page
Dedication
Contents
List of contributors
About the editors
Acknowledgments
1 Microscopy Cancer Cell Imaging in B-lineage Acute Lymphoblastic Leukemia
	1.1 Introduction
	1.2 Building a computer-assisted solution
	1.3 Data preparation
		1.3.1 Preparation of slide for microscopic imaging
		1.3.2 Capture of microscopic images from healthy and cancer subjects for B-acute lymphoblastic leukemia cancer
	1.4 Normalization of color stain to correct for abnormalities during the staining process
		1.4.1 Quantitative results
	1.5 Segmentation of cells of interest (in B-lineage ALL cancer)
		1.5.1 Method-1 of cell segmentation using traditional image processing techniques
		1.5.2 Method-2 of cell segmentation using deep belief network
		1.5.3 Method-3 of cell segmentation using novel convolutional neural network architecture
			1.5.3.1 Brief review of convolutional neural network architectures
			1.5.3.2 Semantic versus instance segmentation in medical imaging
			1.5.3.3 Method-3: novel proposed EDNiS-Net convolutional neural network for automated nuclei instance segmentation
				1.5.3.3.1 Base module
				1.5.3.3.2 Encoder module
				1.5.3.3.3 Decoder module
				1.5.3.3.4 Proposed loss function
				1.5.3.3.5 Results and discussion
			1.5.3.4 Region-proposal based convolutional neural network architectures
	1.6 Classification of cancer and healthy cells
		1.6.1 C-NMC 2019 challenge dataset
		1.6.2 Classification on C-NMC 2019 dataset
		1.6.3 SDCT-AuxNetθ CNN architecture for C-NMC 2019 dataset
	1.7 Conclusions
	References
2 Computational imaging applications in brain and breast cancer
	2.1 Introduction
	2.2 Building upon current clinical standards
		2.2.1 Clinical standards
		2.2.2 Tissue segmentation
	2.3 Deep learning applications in brain cancer
		2.3.1 Tumor grading
		2.3.2 Survival analysis
		2.3.3 Radiogenomics
			2.3.3.1 1p/19q
			2.3.3.2 Isocitrate dehydrogenase
			2.3.3.3 6-methylguanine-DNA methyltransferase
		2.3.4 Pseudoprogression
	2.4 Deep learning applications in breast cancer
		2.4.1 Increasing accuracy in breast cancer risk assessment
		2.4.2 Reproducible breast density assessment for improved breast cancer risk prediction
		2.4.3 Improving performance in breast cancer diagnosis
		2.4.4 Enhancing efficacy in breast cancer clinical practice
	2.5 Conclusion
	Acknowledgments
	References
3 Deep neural networks and advanced computer vision algorithms in the early diagnosis of skin diseases
	3.1 Introduction and motivation for the early diagnosis of melanoma
	3.2 Artificial intelligence and computer vision in melanoma diagnosis
	3.3 Medical diagnostic procedures for screening of skin diseases
	3.4 State-of-the-art survey on skin mole segmentation methods
		3.4.1 Comparison of the state of the art
		3.4.2 Summary
	3.5 Improved local and global patterns detection algorithms by deep learning algorithms
	3.6 Early classification of skin melanomas in dermoscopy
		3.6.1 Diagnostic algorithms
		3.6.2 Approaches to detect the diagnostic criteria
		3.6.3 Approaches to directly classify skin conditions
			3.6.3.1 Classifiers utilizing the convolutional neural networks as a feature extractor
			3.6.3.2 Classifiers using end-to-end learning convolutional neural networks model training with transfer learning
			3.6.3.3 Convolutional neural networks model training from scratch
			3.6.3.4 Ensembles of convolutional neural networks models
	3.7 Conclusions
	3.8 How to speed up the classification process with field-programmable gate arrays?
	3.9 Challenges and future directions
	3.10 Teledermatology
	References
4 An accurate deep learning-based computer-aided diagnosis system for early diagnosis of prostate cancer
	4.1 Introduction
	4.2 Methods
		4.2.1 Feature Extraction
		4.2.2 CNN-based classification
	4.3 Experimental results
	4.4 Conclusion
	References
5 Adaptive graph convolutional neural network and its biomedical applications
	5.1 Introduction
	5.2 Related work
		5.2.1 Evolution of graph convolutional neural networks
			5.2.1.1 Spatial graph convolutional neural networks
			5.2.1.2 Spectral graph convolutional neural networks
		5.2.2 Neural network on molecular graph
		5.2.3 Attention on graph
		5.2.4 Neural network for survival analysis
	5.3 Method
		5.3.1 Spectral graph convolution-LL layer
			5.3.1.1 Learning residual graph Laplacian
			5.3.1.2 Re-parameterization on feature transform
		5.3.2 Adaptive graph convolution network architecture
		5.3.3 Graph attention network on adaptive graph
		5.3.4 DeepGraphSurv framework
	5.4 Experiment
		5.4.1 Drug-property prediction
			5.4.1.1 Baseline model
			5.4.1.2 Dataset
			5.4.1.3 Experimental result
		5.4.2 DeepGraphSurv and survival prediction
			5.4.2.1 Dataset
			5.4.2.2 Baseline model
			5.4.2.3 Experimental result
	5.5 Conclusion
	References
	Further reading
6 Deep slice interpolation via marginal super-resolution, fusion, and refinement
	6.1 Introduction
	6.2 Related work
		6.2.1 Traditional slice interpolation methods
		6.2.2 Learning-based super-resolution methods
	6.3 Problem formulation and baseline convolutional neural networks approaches
	6.4 The proposed algorithm
		6.4.1 Marginal super-resolution
		6.4.2 Two-view fusion and refinement
		6.4.3 Comparison with baseline convolutional neural networks approaches
	6.5 Experiments
		6.5.1 Implementation details
		6.5.2 Dataset
		6.5.3 Evaluation metrics
		6.5.4 Visual comparisons
		6.5.5 Ablation study
	6.6 Conclusion
	References
7 Explainable deep learning approach to predict chemotherapy effect on breast tumor’s MRI
	7.1 Introduction
	7.2 Materials and developed methods
		7.2.1 Study population
		7.2.2 Magnetic resonance imaging protocol
		7.2.3 Image preprocessing
		7.2.4 Convolution neural network architecture development
	7.3 Results
		7.3.1 Quantitative results
		7.3.2 Qualitative results
	7.4 Discussion
	7.5 Conclusion
	Aknowledgments
	References
8 Deep learning interpretability: measuring the relevance of clinical concepts in convolutional neural networks features
	8.1 Introduction
	8.2 Related work on interpretable artificial intelligence
		8.2.1 Motivations
		8.2.2 Related terminology
		8.2.3 Related work on explainable artificial intelligence
			8.2.3.1 Explainable artificial intelligence for medical applications
			8.2.3.2 Visualization methods and feature attribution
			8.2.3.3 Concept attribution
		8.2.4 Evaluation of explainable artificial intelligence methods
	8.3 Methods
		8.3.1 Retinopathy of prematurity
			8.3.1.1 Relevant background
			8.3.1.2 Dataset for the experiments
			8.3.1.3 Task and classification model
		8.3.2 Concept attribution with regression concept vectors
			8.3.2.1 Identification of the concepts
			8.3.2.2 Computing the regression concept vector
			8.3.2.3 Generating local explanations by conceptual sensitivity
			8.3.2.4 Agglomerating scores for global explanations
	8.4 Experiments and results
		8.4.1 Network performance on the retinopathy of prematurity task
		8.4.2 Results of concept attribution
			8.4.2.1 Identification of the concepts
			8.4.2.2 Computation of the regression concept vectors
			8.4.2.3 Evaluation of the conceptual sensitivities
			8.4.2.4 Global explanations with Br
	8.5 Discussion of the results
	8.6 Conclusions
	Acknowledgments
	References
9 Computational lung sound classification: a review
	9.1 Introduction
	9.2 Data processing
		9.2.1 Audio signal preprocessing
			9.2.1.1 Signal splitting
			9.2.1.2 Noise filtering
			9.2.1.3 Resampling
			9.2.1.4 Amplitude scaling
			9.2.1.5 Segment splitting
			9.2.1.6 Padding
		9.2.2 Feature extraction
			9.2.2.1 Features for conventional classifiers
			9.2.2.2 Time-frequency representations for deep learning
		9.2.3 Data augmentation
			9.2.3.1 Time domain
			9.2.3.2 Time–frequency domain
	9.3 Data modeling
		9.3.1 Machine learning
			9.3.1.1 Conventional classifiers
			9.3.1.2 Deep learning architectures
				9.3.1.2.1 Convolutional neural networks
				9.3.1.2.2 Recurrent networks
				9.3.1.2.3 Hybrid systems
		9.3.2 Learning paradigm
			9.3.2.1 Transfer learning
			9.3.2.2 Postprocessing
	9.4 Recent public lung sound datasets
		9.4.1 ICBHI 2017 dataset
		9.4.2 The Abdullah University Hospital 2020 dataset
		9.4.3 HF_Lung_V1 dataset
	9.5 Conclusion
	References
10 Clinical applications of machine learning in heart failure
	10.1 Introduction
	10.2 Diagnosis
		10.2.1 Automatic diagnosis, classification, and phenotyping of heart failure
		10.2.2 Detection of heart failure-associated arrhythmia
	10.3 Management
		10.3.1 Prognostic prediction
		10.3.2 Development of therapy
		10.3.3 Optimal patient selection for specific therapies or recommendation of optimal therapy
	10.4 Prevention
	10.5 Conclusion
	References
11 Role of artificial intelligence and radiomics in diagnosing renal tumors: a survey
	11.1 Introduction
	11.2 Basic background
		11.2.1 Deep learning
		11.2.2 Machine learning
		11.2.3 Radiomics
	11.3 Steps of artificial intelligence-based diagnostic systems
		11.3.1 Image acquisition
		11.3.2 Image segmentation
		11.3.3 Feature extraction and qualifications
		11.3.4 Diagnostic analysis
	11.4 Texture analysis
		11.4.1 Principles
		11.4.2 Statistical techniques
			11.4.2.1 First-order statics
			11.4.2.2 Second-order statics
		11.4.3 Model-based methods
		11.4.4 Transform methods
		11.4.5 Texture parameters
			11.4.5.1 Filtration-histogram method
			11.4.5.2 Postprocessing software
	11.5 Clinical applications of artificial intelligence and radiomics
		11.5.1 Benign versus malignant renal tumors
		11.5.2 Renal cell carcinoma versus angiomyolipoma
		11.5.3 Renal cell carcinoma versus oncocytoma
		11.5.4 Renal cell carcinoma versus renal cyst
		11.5.5 Subtyping of renal cell carcinoma
		11.5.6 Grading of renal cell carcinoma
		11.5.7 Staging of renal cell carcinoma
		11.5.8 Characterization of small renal mass
	11.6 Merits and limitations
		11.6.1 Merits
		11.6.2 Limitations
	11.7 Future directions
	11.8 Conclusion
	References
12 A review of texture-centric diagnostic models for thyroid cancer using convolutional neural networks and visualized text...
	12.1 Introduction
	12.2 Materials and collection protocols
		12.2.1 Study participants and raw data collection
		12.2.2 Nodule segmentation and apparent diffusion coefficient calculations
	12.3 Statistical analysis
	12.4 2D texture model
	12.5 3D texture model
	12.6 Texture analysis
	12.7 Results
		12.7.1 Statistical results
		12.7.2 Diagnostic accuracy of 2D model
			12.7.2.1 Ablation study
			12.7.2.2 Comparison with hand-crafted-based techniques
		12.7.3 Diagnostic accuracy of 3D model
		12.7.4 Texture pattern visualization
	12.8 Discussion
	12.9 Conclusion
	References
Index
Back Cover