Diagnostic Biomedical Signal and Image Processing Applications With Deep Learning Methods: With Deep Learning Methods

Front Cover
Diagnostic Biomedical Signal and Image Processing Applications With Deep Learning Methods
Copyright Page
Contents
List of contributors
1 Introduction to deep learning and diagnosis in medicine
Introduction
Deep learning architectures
Convolutional neural network
AlexNet
ZFNet
NiN
VGGNet
Inception (GoogLeNet)
ResNet
DenseNet
U-Net
SegNet
R-CNN
YOLO
Other convolutional neural networks algorithms
Recurrent neural network
Long short-term memory
Gated recurrent unit
Bidirectional recurrent neural network
Boltzmann machine and restricted Boltzmann machines
Autoencoder
Generative adversarial network
Semisupervised GAN, bidirectional GAN
Conditional GAN, InfoGAN, AC-GAN
LAPGAN, DCGAN, BEGAN
SAGAN, BigGAN
WGAN, WGAN-GP, LSGAN
PROGAN, StyleGAN, StyleGAN2
Comparisons of some GAN models
Other architectures
Deep belief network
Capsule network
Hybrid architectures
Application fields of deep learning in medicine
Clinical and medical images
Biosignals
Biomedicine
Electronic health records
Other fields
Conclusions
References
2 One-dimensional convolutional neural network-based identification of sleep disorders using electroencephalogram signals
Introduction
Materials and methods
Dataset
Method
Results
Discussions
Conclusions
References
3 Classification of histopathological colon cancer images using particle swarm optimization-based feature selection algorithm
Introduction
Methodology
Dataset preparation
Data preprocess and feature extraction
Data size reduction
Global feature extraction
Classifier
Gradient boosting
Feature selection
Particle swarm optimization
Performance metrics
Results
Classification results
Models complexity comparison
SHAP analysis
Receiver operator characteristic analysis
Comparison
Discussion
Conclusion
References
4 Arrhythmia diagnosis from ECG signal pulses with one-dimensional convolutional neural networks
Introduction
Definition of problem
Materials and methods
Dataset
Oversampling
1D-CNN architecture
1D convolution layer
Pooling layer
Batch normalization and dropout layers
Experimental result
Performance metrics
Experimental environment
Random forest classifier
1D-CNN VGG16 classifier results
Discussion
Conclusion and future direction
References
5 Patch-based approaches to whole slide histologic grading of breast cancer using convolutional neural networks
Introduction and motivation
Tubular formation
Nuclear pleomorphism
Mitotic figure detection and classification
Challenges in obtaining Nottingham grading score
Challenges in nuclear pleomorphism classification
Challenges in detection/segmentation of tubular formation
Challenges in mitotic classification
Literature review and state of the art
AI-based approaches for nuclear pleomorphism classification
AI-based approaches for detection and segmentation of tubular formation
AI-based approaches for mitotic classification and counting
Problem/system/application definition
Problem definition and description
System and application definition
Proposed methodology
Pre-processing
Deep learning methods
Mitosis detection and classification
Tubule segmentation
Pleomorphism classification
Results and discussions
Dataset
Assessment
Quantitative assessment
Qualitative assessment
Conclusions
Future work
References
6 Deep neural architecture for breast cancer detection from medical CT image modalities
Introduction
Related work
Experimental work
Dataset
Work flow
Image pre-processing and augmentation methods
Models explored
Experimental results
Evaluation parameters
Models performance on BreakHis dataset
Models performance on BACH2018 dataset
Conclusion
References
7 Automated analysis of phase-contrast optical microscopy time-lapse images: application to wound healing and cell motility...
Introduction and motivation
Literature review and state of the art
Pre-processing of PCM time-lapse images
Segmentation of PCM time-lapse images
Tracking and quantification from PCM time-lapse images
Workflows for the analysis of PCM time-lapse images
Problem definition, acquisition and annotation of data
Data acquisition
Data annotation
Proposed solution
Pre-processing
Segmentation
Tracking and quantification
Qualitative and quantitative analysis
Pre-processing
Segmentation
Tracking and quantification
Use cases and applications
Discussion
Conclusions
Outlook and future work
Software availability
Acknowledgment
References
8 Automatic detection of pathological changes in chest X-ray screening images using deep learning methods
Introduction
Screening for lung abnormalities
Introduction
Original image data
Normal cases
Pathological cases
Image data preprocessing
Methods
Results
Local conclusions
Detecting extrapulmonary pathologies
Introduction
Data preparation
Computational experiment
Local conclusions
Identification of subjects with lung roots abnormalities
Introduction
Materials
Methods
Results
Local conclusions
Chest X-ray image analysis web services
Overview
Authentication
Authentication
Input data validation
X-ray modality checker for 2D images
Anatomy checker for 2D images
Axes order and orientation checker for 2D chest X-ray images
Resources management
Applications for processing and analyzing chest X-ray
Lung segmentation on chest X-rays
Detecting abnormalities in chest X-rays (heatmap)
Application for computer-aided diagnostics based on chest X-ray
Conclusion
References
9 Dependence of the results of adversarial attacks on medical image modality, attack type, and defense methods
Introduction
Materials
Chest X-ray images
CT images
Histopathology images
Methods
Attacks
FGSM Attacks
AutoAttacks
Carlini-Wagner Attacks
Defenses
Adversarial training
High-level representation guided denoiser
The MagNet
Experimental pipeline
Results
Experiments with X-ray images
Experiments with computer tomography images
Experiments with histopathology images
Discussion
The abilities of adversarial training defense method
Important properties of class-label-guided denoiser defense
Important properties of MagNet defense
Conclusions
References
10 A deep ensemble network for lung segmentation with stochastic weighted averaging
Introduction
Related works
Proposed system
Dataset collection
Data augmentation
Segmentation architectures
HarDNet
UNet++
Deeplab V3—ResNet
Stochastic weighted averaging (SWA)
Ensemble
Results and discussion
Dataset description
Ablation studies
Analysis of HarDNet
Analysis of UNet++
Analysis of ResNet
Analysis of ensemble
Performance analysis
Conclusion
References
11 Deep ensembles and data augmentation for semantic segmentation
Introduction
Methods
Deep learning for semantic image segmentation
Loss functions
Dice Loss
Tversky Loss
Focal Tversky Loss
Focal Generalized Dice Loss
Log-Cosh Type Losses
SSIM Loss
Different functions combined loss
Data augmentation
Shadows
Contrast and motion blur
Color mapping
Experimental results
Metrics
Datasets and testing protocol for polyp segmentation
Datasets and testing protocol for skin segmentation
Datasets and testing protocol for leukocyte segmentation
Experiments
Conclusions
Acknowledgment
References
12 Classification of diseases from CT images using LSTM-based CNN
Introduction
Background
CT dataset-issues and challenges in handling them
Elucidating classical CNN- and LSTM-based CNN models
Convolutional neural network
Convolution layer
Pooling layer
Fully connected layers
LSTM networks
Previous work done on CNN-LSTM
Conclusion
References
13 A novel polyp segmentation approach using U-net with saliency-like feature fusion
Introduction
Methodology
Image enhancement
Discriminatory feature matrices
Fusion of feature matrices
U-net fine-tuning
Loss function
Experiments and results
Datasets
Evaluation metrics
Experimental results of enhanced images with image inpainting method
Experimental results of proposed method
Discussion
Conclusion
Compliance with ethical standards
Conflict of interest
Human and animal rights
References
Index
Back Cover