Schweitzer Fachinformationen
Wenn es um professionelles Wissen geht, ist Schweitzer Fachinformationen wegweisend. Kunden aus Recht und Beratung sowie Unternehmen, öffentliche Verwaltungen und Bibliotheken erhalten komplette Lösungen zum Beschaffen, Verwalten und Nutzen von digitalen und gedruckten Medien.
The book details deep learning models like ANN, RNN, LSTM, in many industrial sectors such as transportation, healthcare, military, agriculture, with valid and effective results, which will help researchers find solutions to their deep learning research problems.
We have entered the era of smart world devices, where robots or machines are being used in most applications to solve real-world problems. These smart machines/devices reduce the burden on doctors, which in turn make their lives easier and the lives of their patients better, thereby increasing patient longevity, which is the ultimate goal of computer vision. Therefore, the goal in writing this book is to attempt to provide complete information on reliable deep learning models required for e-healthcare applications. Ways in which deep learning can enhance healthcare images or text data for making useful decisions are discussed. Also presented are reliable deep learning models, such as neural networks, convolutional neural networks, backpropagation, and recurrent neural networks, which are increasingly being used in medical image processing, including for colorization of black and white X-ray images, automatic machine translation images, object classification in photographs/images (CT scans), character or useful generation (ECG), image caption generation, etc. Hence, reliable deep learning methods for the perception or production of better results are a necessity for highly effective e-healthcare applications. Currently, the most difficult data-related problem that needs to be solved concerns the rapid increase of data occurring each day via billions of smart devices. To address the growing amount of data in healthcare applications, challenges such as not having standard tools, efficient algorithms, and a sufficient number of skilled data scientists need to be overcome. Hence, there is growing interest in investigating deep learning models and their use in e-healthcare applications.
Audience Researchers in artificial intelligence, big data, computer science, and electronic engineering, as well as industry engineers in transportation, healthcare, biomedicine, military, agriculture.
Amit Kumar Tyagi is an assistant professor and senior researcher at Vellore Institute of Technology (VIT), Chennai Campus, India. He received his PhD in 2018 from Pondicherry Central University, India. He has published more than 8 patents in the area of deep learning, Internet of Things, cyber physical systems and computer vision.
Preface xix
Part I: Deep Learning and Its Models 1
1 CNN: A Review of Models, Application of IVD Segmentation 3Leena Silvoster M. and R. Mathusoothana S. Kumar
1.1 Introduction 4
1.2 Various CNN Models 4
1.2.1 LeNet-5 4
1.2.2 AlexNet 7
1.2.3 ZFNet 8
1.2.4 VGGNet 10
1.2.5 GoogLeNet 12
1.2.6 ResNet 16
1.2.7 ResNeXt 21
1.2.8 SE-ResNet 24
1.2.9 DenseNet 24
1.2.10 MobileNets 25
1.3 Application of CNN to IVD Detection 26
1.4 Comparison With State-of-the-Art Segmentation Approaches for Spine T2W Images 28
1.5 Conclusion 28
References 33
2 Location-Aware Keyword Query Suggestion Techniques With Artificial Intelligence Perspective 35R. Ravinder Reddy, C. Vaishnavi, Ch. Mamatha and S. Ananthakumaran
2.1 Introduction 36
2.2 Related Work 39
2.3 Artificial Intelligence Perspective 41
2.3.1 Keyword Query Suggestion 42
2.3.1.1 Random Walk-Based Approaches 42
2.3.1.2 Cluster-Based Approaches 42
2.3.1.3 Learning to Rank Approaches 43
2.3.2 User Preference From Log 43
2.3.3 Location-Aware Keyword Query Suggestion 44
2.3.4 Enhancement With AI Perspective 44
2.3.4.1 Case Study 45
2.4 Architecture 46
2.4.1 Distance Measures 47
2.5 Conclusion 49
References 49
3 Identification of a Suitable Transfer Learning Architecture for Classification: A Case Study with Liver Tumors 53B. Lakshmi Priya, K. Jayanthi, Biju Pottakkat and G. Ramkumar
3.1 Introduction 54
3.2 Related Works 56
3.3 Convolutional Neural Networks 58
3.3.1 Feature Learning in CNNs 59
3.3.2 Classification in CNNs 60
3.4 Transfer Learning 61
3.4.1 AlexNet 61
3.4.2 GoogLeNet 62
3.4.3 Residual Networks 63
3.4.3.1 ResNet-18 65
3.4.3.2 ResNet-50 65
3.5 System Model 66
3.6 Results and Discussions 67
3.6.1 Dataset 67
3.6.2 Assessment of Transfer Learning Architectures 67
3.7 Conclusion 73
References 74
4 Optimization and Deep Learning-Based Content Retrieval, Indexing, and Metric Learning Approach for Medical Images 79Suresh Kumar K., Sundaresan S., Nishanth R. and Ananth Kumar T.
4.1 Introduction 80
4.2 Related Works 82
4.3 Proposed Method 85
4.3.1 Input Dataset 86
4.3.2 Pre-Processing 86
4.3.3 Combination of DCNN and CFML 86
4.3.4 Fine Tuning and Optimization 88
4.3.5 Feature Extraction 89
4.3.6 Localization of Abnormalities in MRI and CT Scanned Images 90
4.4 Results and Discussion 92
4.4.1 Metric Learning 92
4.4.2 Comparison of the Various Models for Image Retrieval 92
4.4.3 Precision vs. Recall Parameters Estimation for the CBIR 93
4.4.4 Convolutional Neural Networks-Based Landmark Localization 96
4.5 Conclusion 104
References 104
Part II: Applications of Deep Learning 107
5 Deep Learning for Clinical and Health Informatics 109Amit Kumar Tyagi and Meghna Mannoj Nair
5.1 Introduction 110
5.1.1 Deep Learning Over Machine Learning 111
5.2 Related Work 113
5.3 Motivation 115
5.4 Scope of the Work in Past, Present, and Future 115
5.5 Deep Learning Tools, Methods Available for Clinical, and Health Informatics 117
5.6 Deep Learning: Not-So-Near Future in Biomedical Imaging 119
5.6.1 Types of Medical Imaging 119
5.6.2 Use and Benefits of Medical Imaging 120
5.7 Challenges Faced Toward Deep Learning Using Biomedical Imaging 121
5.7.1 Deep Learning in Healthcare: Limitations and Challenges 122
5.8 Open Research Issues and Future Research Directions Biomedical Imaging (Healthcare Informatics) 124
5.9 Conclusion 127
References 127
6 Biomedical Image Segmentation by Deep Learning Methods 131K. Anita Davamani, C.R. Rene Robin, S. Amudha and L. Jani Anbarasi
6.1 Introduction 132
6.2 Overview of Deep Learning Algorithms 135
6.2.1 Deep Learning Classifier (DLC) 136
6.2.2 Deep Learning Architecture 137
6.3 Other Deep Learning Architecture 139
6.3.1 Restricted Boltzmann Machine (RBM) 139
6.3.2 Deep Learning Architecture Containing Autoencoders 140
6.3.3 Sparse Coding Deep Learning Architecture 141
6.3.4 Generative Adversarial Network (GAN) 141
6.3.5 Recurrent Neural Network (RNN) 141
6.4 Biomedical Image Segmentation 145
6.4.1 Clinical Images 146
6.4.2 X-Ray Imaging 146
6.4.3 Computed Tomography (CT) 147
6.4.4 Magnetic Resonance Imaging (MRI) 147
6.4.5 Ultrasound Imaging (US) 148
6.4.6 Optical Coherence Tomography (OCT) 148
6.5 Conclusion 149
References 149
7 Multi-Lingual Handwritten Character Recognition Using Deep Learning 155Giriraj Parihar, Ratnavel Rajalakshmi and Bhuvana J.
7.1 Introduction 156
7.2 Related Works 157
7.3 Materials and Methods 160
7.4 Experiments and Results 161
7.4.1 Dataset Description 162
7.4.1.1 Handwritten Math Symbols 162
7.4.1.2 Bangla Handwritten Character Dataset 162
7.4.1.3 Devanagari Handwritten Character Dataset 162
7.4.2 Experimental Setup 162
7.4.3 Hype-Parameters 164
7.4.3.1 English Model 164
7.4.3.2 Hindi Model 165
7.4.3.3 Bangla Model 165
7.4.3.4 Math Symbol Model 165
7.4.3.5 Combined Model 166
7.4.4 Results and Discussion 167
7.4.4.1 Performance of Uni-Language Models 167
7.4.4.2 Uni-Language Model on English Dataset 168
7.4.4.3 Uni-Language Model on Hindi Dataset 168
7.4.4.4 Uni-Language Model on Bangla Dataset 169
7.4.4.5 Uni-Language Model on Math Symbol Dataset 169
7.4.4.6 Performance of Multi-Lingual Model on Combined Dataset 171
7.5 Conclusion 177
References 178
8 Disease Detection Platform Using Image Processing Through OpenCV 181Neetu Faujdar and Aparna Sinha
8.1 Introduction 182
8.1.1 Image Processing 183
8.2 Problem Statement 183
8.2.1 Cataract 183
8.2.1.1 Causes 184
8.2.1.2 Types of Cataracts 184
8.2.1.3 Cataract Detection 185
8.2.1.4 Treatment 186
8.2.1.5 Prevention 186
8.2.1.6 Methodology 186
8.2.2 Eye Cancer 192
8.2.2.1 Symptoms 194
8.2.2.2 Causes of Retinoblastoma 194
8.2.2.3 Phases 195
8.2.2.4 Spreading of Cancer 196
8.2.2.5 Diagnosis 196
8.2.2.6 Treatment 197
8.2.2.7 Methodology 199
8.2.3 Skin Cancer (Melanoma) 202
8.2.3.1 Signs and Symptoms 203
8.2.3.2 Stages 203
8.2.3.3 Causes of Melanoma 204
8.2.3.4 Diagnosis 204
8.2.3.5 Treatment 205
8.2.3.6 Methodology 206
8.2.3.7 Asymmetry 207
8.2.3.8 Border 208
8.2.3.9 Color 208
8.2.3.10 Diameter Detection 209
8.2.3.11 Calculating TDS (Total Dermoscopy Score) 210
8.3 Conclusion 210
8.4 Summary 212
References 212
9 Computer-Aided Diagnosis of Liver Fibrosis in Hepatitis Patients Using Convolutional Neural Network 217Aswathy S. U., Ajesh F., Shermin Shamsudheen and Jarin T.
9.1 Introduction 218
9.2 Overview of System 219
9.3 Methodology 219
9.3.1 Dataset 220
9.3.2 Pre-Processing 221
9.3.3 Feature Extraction 221
9.3.4 Feature Selection and Normalization 223
9.3.5 Classification Model 225
9.4 Performance and Analysis 227
9.5 Experimental Results 232
9.6 Conclusion and Future Scope 232
References 233
Part III: Future Deep Learning Models 237
10 Lung Cancer Prediction in Deep Learning Perspective 239Nikita Banerjee and Subhalaxmi Das
10.1 Introduction 239
10.2 Machine Learning and Its Application 240
10.2.1 Machine Learning 240
10.2.2 Different Machine Learning Techniques 241
10.2.2.1 Decision Tree 242
10.2.2.2 Support Vector Machine 242
10.2.2.3 Random Forest 242
10.2.2.4 K-Means Clustering 242
10.3 Related Work 243
10.4 Why Deep Learning on Top of Machine Learning? 245
10.4.1 Deep Neural Network 246
10.4.2 Deep Belief Network 247
10.4.3 Convolutional Neural Network 247
10.5 How is Deep Learning Used for Prediction of Lungs Cancer? 248
10.5.1 Proposed Architecture 248
10.5.1.1 Pre-Processing Block 250
10.5.1.2 Segmentation 250
10.5.1.3 Classification 252
10.6 Conclusion 253
References 253
11 Lesion Detection and Classification for Breast Cancer Diagnosis Based on Deep CNNs from Digital Mammographic Data 257Diksha Rajpal, Sumita Mishra and Anil Kumar
11.1 Introduction 257
11.2 Background 258
11.2.1 Methods of Diagnosis of Breast Cancer 258
11.2.2 Types of Breast Cancer 260
11.2.3 Breast Cancer Treatment Options 261
11.2.4 Limitations and Risks of Diagnosis and Treatment Options 262
11.2.4.1 Limitation of Diagnosis Methods 262
11.2.4.2 Limitations of Treatment Plans 263
11.2.5 Deep Learning Methods for Medical Image Analysis: Tumor Classification 263
11.3 Methods 265
11.3.1 Digital Repositories 265
11.3.1.1 DDSM Database 265
11.3.1.2 AMDI Database 265
11.3.1.3 IRMA Database 265
11.3.1.4 BreakHis Database 265
11.3.1.5 MIAS Database 266
11.3.2 Data Pre-Processing 266
11.3.2.1 Advantages of Pre-Processing Images 267
11.3.3 Convolutional Neural Networks (CNNs) 268
11.3.3.1 Architecture of CNN 269
11.3.4 Hyper-Parameters 272
11.3.4.1 Number of Hidden Layers 273
11.3.4.2 Dropout Rate 273
11.3.4.3 Activation Function 273
11.3.4.4 Learning Rate 274
11.3.4.5 Number of Epochs 274
11.3.4.6 Batch Size 274
11.3.5 Techniques to Improve CNN Performance 274
11.3.5.1 Hyper-Parameter Tuning 274
11.3.5.2 Augmenting Images 274
11.3.5.3 Managing Over-Fitting and Under-Fitting 275
11.4 Application of Deep CNN for Mammography 275
11.4.1 Lesion Detection and Localization 275
11.4.2 Lesion Classification 279
11.5 System Model and Results 280
11.5.1 System Model 280
11.5.2 System Flowchart 281
11.5.2.1 MIAS Database 281
11.5.2.2 Unannotated Images 281
11.5.3 Results 282
11.5.3.1 Distribution and Processing of Dataset 282
11.5.3.2 Training of the Model 283
11.5.3.3 Prediction of Unannotated Images 286
11.6 Research Challenges and Discussion on Future Directions 286
11.7 Conclusion 288
References 289
12 Health Prediction Analytics Using Deep Learning Methods and Applications 293Sapna Jain, M. Afshar Alam, Nevine Makrim Labib and Eiad Yafi
12.1 Introduction 294
12.2 Background 298
12.3 Predictive Analytics 299
12.4 Deep Learning Predictive Analysis Applications 305
12.4.1 Deep Learning Application Model to Predict COVID-19 Infection 305
12.4.2 Deep Transfer Learning for Mitigating the COVID-19 Pandemic 308
12.4.3 Health Status Prediction for the Elderly Based on Machine Learning 309
12.4.4 Deep Learning in Machine Health Monitoring 311
12.5 Discussion 319
12.6 Conclusion 320
References 321
13 Ambient-Assisted Living of Disabled Elderly in an Intelligent Home Using Behavior Prediction-A Reliable Deep Learning Prediction System 329Sophia S., Sridevi U.K., Boselin Prabhu S.R. and P. Thamaraiselvi
13.1 Introduction 330
13.2 Activities of Daily Living and Behavior Analysis 331
13.3 Intelligent Home Architecture 333
13.4 Methodology 335
13.4.1 Record the Behaviors Using Sensor Data 335
13.4.2 Classify Discrete Events and Relate the Events Using Data Analysis Algorithms 335
13.4.3 Construct Behavior Dictionaries for Flexible Event Intervals Using Deep Learning Concepts 335
13.4.4 Use the Dictionary in Modeling the Behavior Patterns Through Prediction Techniques 336
13.4.5 Detection of Deviations From Expected Behaviors Aiding the Automated Elderly Monitoring Based on Decision Support Algorithm Systems 336
13.5 Senior Analytics Care Model 337
13.6 Results and Discussions 338
13.7 Conclusion 341
Nomenclature 341
References 342
14 Early Diagnosis Tool for Alzheimer's Disease Using 3D Slicer 343V. Krishna Kumar, M.S. Geetha Devasena and G. Gopu
14.1 Introduction 344
14.2 Related Work 345
14.3 Existing System 347
14.4 Proposed System 347
14.4.1 Usage of 3D Slicer 350
14.5 Results and Discussion 353
14.6 Conclusion 356
References 356
Part IV: Deep Learning - Importance and Challenges for Other Sectors 361
15 Deep Learning for Medical Healthcare: Issues, Challenges, and Opportunities 363Meenu Gupta, Akash Gupta and Gaganjot Kaur
15.1 Introduction 364
15.2 Related Work 365
15.3 Development of Personalized Medicine Using Deep Learning: A New Revolution in Healthcare Industry 367
15.3.1 Deep Feedforward Neural Network (DFF) 367
15.3.2 Convolutional Neural Network 367
15.3.3 Recurrent Neural Network (RNN) 369
15.3.4 Long/Short-Term Memory (LSTM) 369
15.3.5 Deep Belief Network (DBN) 370
15.3.6 Autoencoder (AE) 370
15.4 Deep Learning Applications in Precision Medicine 370
15.4.1 Discovery of Biomarker and Classification of Patient 370
15.4.2 Medical Imaging 371
15.5 Deep Learning for Medical Imaging 372
15.5.1 Medical Image Detection 372
15.5.1.1 Pathology Detection 372
15.5.1.2 Detection of Image Plane 373
15.5.1.3 Anatomical Landmark Localization 373
15.5.2 Medical Image Segmentation 373
15.5.2.1 Supervised Algorithms 374
15.5.2.2 Semi-Supervised Algorithms 374
15.5.3 Medical Image Enhancement 375
15.5.3.1 Two-Dimensional Super-Resolution Techniques 375
15.5.3.2 Three-Dimensional Super-Resolution Techniques 375
15.6 Drug Discovery and Development: A Promise Fulfilled by Deep Learning Technology 375
15.6.1 Prediction of Drug Properties 376
15.6.2 Prediction of Drug-Target Interaction 377
15.7 Application Areas of Deep Learning in Healthcare 377
15.7.1 Medical Chatbots 377
15.7.2 Smart Health Records 377
15.7.3 Cancer Diagnosis 378
15.8 Privacy Issues Arising With the Usage of Deep Learning in Healthcare 379
15.8.1 Private Data 379
15.8.2 Privacy Attacks 380
15.8.2.1 Evasion Attack 380
15.8.2.2 White-Box Attack 380
15.8.2.3 Black-Box Attack 380
15.8.2.4 Poisoning Attack 381
15.8.3 Privacy-Preserving Techniques 381
15.8.3.1 Differential Privacy With Deep Learning 381
15.8.3.2 Homomorphic Encryption (HE) on Deep Learning 382
15.8.3.3 Secure Multiparty Computation on Deep Learning 383
15.9 Challenges and Opportunities in Healthcare Using Deep Learning 383
15.10 Conclusion and Future Scope 386
References 387
16 A Perspective Analysis of Regularization and Optimization Techniques in Machine Learning 393Ajeet K. Jain, PVRD Prasad Rao and K. Venkatesh Sharma
16.1 Introduction 394
16.1.1 Data Formats 395
16.1.1.1 Structured Data 395
16.1.1.2 Unstructured Data 396
16.1.1.3 Semi-Structured Data 396
16.1.2 Beginning With Learning Machines 397
16.1.2.1 Perception 397
16.1.2.2 Artificial Neural Network 398
16.1.2.3 Deep Networks and Learning 399
16.1.2.4 Model Selection, Over-Fitting, and Under-Fitting 400
16.2 Regularization in Machine Learning 402
16.2.1 Hamadard Conditions 403
16.2.2 Tikhonov Generalized Regularization 404
16.2.3 Ridge Regression 406
16.2.4 Lasso-L1 Regularization 406
16.2.5 Dropout as Regularization Feature 407
16.2.6 Augmenting Dataset 408
16.2.7 Early Stopping Criteria 408
16.3 Convexity Principles 409
16.3.1 Convex Sets 410
16.3.1.1 Affine Set and Convex Functions 411
16.3.1.2 Properties of Convex Functions 411
16.3.2 Optimization and Role of Optimizer in ML 413
16.3.2.1 Gradients-Descent Optimization Methods 414
16.3.2.2 Non-Convexity of Cost Functions 416
16.3.2.3 Basic Maths of SGD 418
16.3.2.4 Saddle Points 418
16.3.2.5 Gradient Pointing in the Wrong Direction 420
16.3.2.6 Momentum-Based Optimization 423
16.4 Conclusion and Discussion 424
References 425
17 Deep Learning-Based Prediction Techniques for Medical Care: Opportunities and Challenges 429S. Subasree and N. K. Sakthivel
17.1 Introduction 430
17.2 Machine Learning and Deep Learning Framework 431
17.2.1 Supervised Learning 433
17.2.2 Unsupervised Learning 433
17.2.3 Reinforcement Learning 434
17.2.4 Deep Learning 434
17.3 Challenges and Opportunities 435
17.3.1 Literature Review 435
17.4 Clinical Databases-Electronic Health Records 436
17.5 Data Analytics Models-Classifiers and Clusters 436
17.5.1 Criteria for Classification 438
17.5.1.1 Probabilistic Classifier 439
17.5.1.2 Support Vector Machines (SVMs) 439
17.5.1.3 K-Nearest Neighbors 440
17.5.2 Criteria for Clustering 441
17.5.2.1 K-Means Clustering 442
17.5.2.2 Mean Shift Clustering 442
17.6 Deep Learning Approaches and Association Predictions 444
17.6.1 G-HR: Gene Signature-Based HRF Cluster 444
17.6.1.1 G-HR Procedure 446
17.6.2 Deep Learning Approach and Association Predictions 446
17.6.2.1 Deep Learning Approach 446
17.6.2.2 Intelligent Human Disease-Gene Association Prediction Technique (IHDGAP) 447
17.6.2.3 Convolution Neural Network 447
17.6.2.4 Disease Semantic Similarity 449
17.6.2.5 Computation of Scoring Matrix 450
17.6.3 Identified Problem 450
17.6.4 Deep Learning-Based Human Diseases Pattern Prediction Technique for High-Dimensional Human Diseases Datasets (ECNN-HDPT) 451
17.6.5 Performance Analysis 453
17.7 Conclusion 457
17.8 Applications 458
References 459
18 Machine Learning and Deep Learning: Open Issues and Future Research Directions for the Next 10 Years 463Akshara Pramod, Harsh Sankar Naicker and Amit Kumar Tyagi
18.1 Introduction 464
18.1.1 Comparison Among Data Mining, Machine Learning, and Deep Learning 465
18.1.2 Machine Learning 465
18.1.2.1 Importance of Machine Learning in Present Business Scenario 467
18.1.2.2 Applications of Machine Learning 467
18.1.2.3 Machine Learning Methods Used in Current Era 469
18.1.3 Deep Learning 471
18.1.3.1 Applications of Deep Learning 471
18.1.3.2 Deep Learning Techniques/Methods Used in Current Era 473
18.2 Evolution of Machine Learning and Deep Learning 475
18.3 The Forefront of Machine Learning Technology 476
18.3.1 Deep Learning 476
18.3.2 Reinforcement Learning 477
18.3.3 Transfer Learning 477
18.3.4 Adversarial Learning 477
18.3.5 Dual Learning 478
18.3.6 Distributed Machine Learning 478
18.3.7 Meta Learning 478
18.4 The Challenges Facing Machine Learning and Deep Learning 478
18.4.1 Explainable Machine Learning 479
18.4.2 Correlation and Causation 479
18.4.3 Machine Understands the Known and is Aware of the Unknown 479
18.4.4 People-Centric Machine Learning Evolution 480
18.4.5 Explainability: Stems From Practical Needs and Evolves Constantly 480
18.5 Possibilities With Machine Learning and Deep Learning 481
18.5.1 Possibilities With Machine Learning 481
18.5.1.1 Lightweight Machine Learning and Edge Computing 481
18.5.1.2 Quantum Machine Learning 482
18.5.1.3 Quantum Machine Learning Algorithms Based on Linear Algebra 482
18.5.1.4 Quantum Reinforcement Learning 483
18.5.1.5 Simple and Elegant Natural Laws 483
18.5.1.6 Improvisational Learning 484
18.5.1.7 Social Machine Learning 485
18.5.2 Possibilities With Deep Learning 485
18.5.2.1 Quantum Deep Learning 485
18.6 Potential Limitations of Machine Learning and Deep Learning 486
18.6.1 Machine Learning 486
18.6.2 Deep Learning 487
18.7 Conclusion 488
Acknowledgement 489
Contribution/Disclosure 489
References 489
Index 491
Leena Silvoster M.1* and R. Mathusoothana S. Kumar2
1Department of Computer Science Engg, College of Engg, Attingal, Thiruvananthapuram, Kerala, India
2Department of Information Technology, Noorul Islam University, Tamilnadu, India
Abstract
The widespread publicity of Convolutional Neural Network (CNN) in various domains such as image classification, object recognition, and scene classification has revolutionized the research in machine learning, especially in medical images. Magnetic Resonance Images (MRIs) are suffering from severe noise, weak edges, low contrast, and intensity inhomogeneity. Recent advances in deep learning with fewer connections and parameters made their training easier. This chapter presents an in-depth review of the various deep architectures as well as its application for segmenting the Intervertebral disc (IVD) from the 3D spine image and its evaluation. The first section deals with the study of various traditional architectures of deep CNN such as LeNet, AlexNet, ZFNet, GoogleNet, VGGNet, ResNet, Inception model, ResNeXt, SENet, MobileNet V1/V2, and DenseNet. It also deals with the study of the parameters and components associated with the models in detail. The second section discusses the application of these models to segment IVD from the spine image. Finally, theoretically performance and experimental results of the state-of-art of the literature shows that 2.5D multi-scale FCN performs the best with the Dice Similarity Index (DSC) of 90.64%.
Keywords: CNN, deep learning, intervertebral disc degeneration, MRI segmentation
The concept of Convolutional Neural Network (CNN) was introduced by Fukushima. The principle in CNN is that the visual mechanism of human is hierarchical in structure. CNN has been successfully applied in various image domain such as image classification, object recognition, and scene classification. CNN is defined as a series of convolution layer and pooling layer. In the convolution layer, the image is convolved with a filter, i.e., slide over the image spatially and computing dot products. Pooling layer provides a smaller feature set.
One major cause of low back pain is disc degeneration. Automated detection of lumbar abnormalities from the clinical scan is a burden for radiologist. Researchers focus on the automation task of the segmentation of large set of MRI data due to the huge size of such images. The success of the application of CNN in various field of object detection enables the researchers to apply various models for the detection of Intervertebral Disc (IVD) and, in turn, helps in the diagnosis of diseases.
The details of the structure of the remaining section of the paper are as follows. The next section deals with the study of the various CNN models. Section 1.3, presents applications of CNN for the detection of the IVD. In Section 1.4, comparison with state-of-the-art segmentation approaches for spine T2W images is carried out, and conclusion is in Section 1.5.
The LeNet architecture was proposed by LeCun et al. [1], and it successfully classified the images in the MNIST dataset. LeNet uses grayscale image of 32×32 pixel as input image. As a pre-processing step the input pixel values are normalized so that white (background) pixel represents a value of 1 and the black (foreground) represents a value of 1.175, which, in turn, speedup the learning task. The LeNet-5 architecture consists of succession of input layer, two sets of convolutional and average pooling layers, followed by a flattening convolutional layer, then two fully connected layers, and finally a softmax classifier.
The first convolutional layer filters the 32×32 input image with six filters. All filter kernels are of size 5×5 (receptive field) with a stride of 1 pixel (this is the distance between the receptive field centers of neighboring neurons in a kernel map) and uses "same" padding. Given the input image of size 28×28, apply six convolutional kernels each of size 5×5 with stride 1 in C1, the feature maps obtained is of size 14×14. Figure 1.1 shows the architecture of LeNet-5, and Table 1.1 shows the various parameter details of LeNet-5. Let Wc is the number of weights in the layer; Bc is the number of biases in the layer; Pc is the number of parameters in the layer; K is the size (width) of kernels in the layer; N is the number of kernels; C is the number of channels in the input image.
In the first convolutional layer, number of learning parameters is (5×5 + 1) × 6 = 156 parameters; where 6 is the number of filters, 5 × 5 is the filter size, and bias is 1, and there are 28×28×156 = 122,304 connections. The number of feature map calculation is as follows:
W = 32; H = 32; Fw = Fh = 5; P = 0, and the number of feature map is 28 × 28.
First pooling layer: W = 28; H = 28; P = 0; S = 2
Figure 1.1 Architecture of LeNet-5.
Table 1.1 Various parameters of the layers of LeNet.
The number of feature map is 14×14 and the number of learning parameters is (coefficient + bias) × no. filters = (1+1) × 6 = 12 parameters and the number of connections = 30×14×14 = 5,880.
Layer 3: In this layer, only 10 out of 16 feature maps are connected to six feature maps of the previous layer as shown in Table 1.2. Each unit in C3 is connected to several 5 × 5 receptive fields at identical locations in S2. Total number of trainable parameters = (3×5×5+1)×6+(4×5×5+1)×9+(6×5×5+1) = 1516. Total number of connections = (3×5×5+1)×6×10×10+(4×5×5+1) ×9×10×10 +(6×5×5+1)×10×10 = 151,600. Total number of parameters is 60K.
Alex Krizhevsky et al. [2] presented a new architecture "AlexNet" to train the ImageNet dataset, which consists of 1.2 million high-resolution images, into 1,000 different classes. In the original implementation, layers are divided into two and to train them on separate GPUs (GTX 580 3GB GPUs) takes around 5-6 days. The network contains five convolutional layers, maximum pooling layers and it is followed by three fully connected layers, and finally a 1,000-way softmax classifier. The network uses ReLU activation function, data augmentation, dropout and smart optimizer layers, local response normalization, and overlapping pooling. The AlexNet has 60M parameters. Figure 1.2 shows the architecture of AlexNet and Table 1.3 shows the various parameters of AlexNet.
Table 1.2 Every column indicates which feature map in S2 are combined by the units in a particular feature map of C3...
Dateiformat: ePUBKopierschutz: Adobe-DRM (Digital Rights Management)
Systemvoraussetzungen:
Das Dateiformat ePUB ist sehr gut für Romane und Sachbücher geeignet – also für „fließenden” Text ohne komplexes Layout. Bei E-Readern oder Smartphones passt sich der Zeilen- und Seitenumbruch automatisch den kleinen Displays an. Mit Adobe-DRM wird hier ein „harter” Kopierschutz verwendet. Wenn die notwendigen Voraussetzungen nicht vorliegen, können Sie das E-Book leider nicht öffnen. Daher müssen Sie bereits vor dem Download Ihre Lese-Hardware vorbereiten.Bitte beachten Sie: Wir empfehlen Ihnen unbedingt nach Installation der Lese-Software diese mit Ihrer persönlichen Adobe-ID zu autorisieren!
Weitere Informationen finden Sie in unserer E-Book Hilfe.
