Signal Processing in Medicine and Biology : Emerging Trends in Research and Applications

This book covers emerging trends in signal processing research and biomedical engineering, exploring the ways in which signal processing plays a vital role in applications ranging from medical electronics to data mining of electronic medical records. Topics covered include statistical modeling of electroencephalograph data for predicting or detecting seizure, stroke, or Parkinson's; machine learning methods and their application to biomedical problems, which is often poorly understood, even within the scientific community; signal analysis; medical imaging; and machine learning, data mining, and classification. The book features tutorials and examples of successful applications that will appeal to a wide range of professionals and researchers interested in applications of signal processing, medicine, and biology. Covers traditional signal processing topics within biomedicine Promotes collaboration between healthcare practitioners and signal processing researchers Presents tutorials and examples of successful applications Preface Contents 1 An Analysis of Automated Parkinson's Diagnosis Using Voice: Methodology and Future Directions 1.1 Introduction 1.1.1 Voice as a Biomarker 1.1.2 Parkinson's Disease Background 1.1.3 Parkinson's Disease Detection—Current Methods 1.1.4 Parkinson's Disease Pathophysiology 1.1.5 Previous Work in Parkinson's Diagnosis Using Voice 1.2 mPower Voice Dataset 1.3 Methods 1.3.1 Voice Activation Detection 1.3.2 Feature Selection 1.3.2.1 Mel-Frequency Cepstrum Coefficients 1.3.2.2 GeMAPS Features 1.3.2.3 Audio Visual Emotion Recognition Challenge 2013 (AVEC) Features 1.3.3 Maximum Relevance Minimum Redundancy 1.3.4 Machine Learning 1.3.4.1 Cross Validation and Grid Search 1.3.4.2 Decision Trees 1.3.4.3 Random Forest 1.3.4.4 Extra Trees 1.3.4.5 Gradient Boosted Decision Trees 1.3.4.6 Support Vector Machine 1.3.4.7 Artificial Neural Networks 1.4 Results 1.5 Discussion 1.6 Conclusion References 2 Noninvasive Vascular Blood Sound Monitoring Through Flexible Microphone 2.1 Introduction and Background 2.2 Prior Work in Phonoangiographic Detection of Stenosis 2.2.1 Effect of Stenosis and Blood Flow on Bruit Spectra and Intensity 2.2.2 Effect of Recording Location on Bruit Spectra 2.2.3 Stenosis Severity Classification from PAG Signal Analysis 2.3 Phonoangiogram Signal Processing 2.3.1 Bruit-Enhancing Filter 2.3.2 PAG Wavelet Analysis 2.3.3 Wavelet-Derived Auditory Signals 2.3.4 PAG Systole/Diastole Segmentation 2.3.5 PAG Spectral Feature Extraction 2.4 Skin-Coupled Recording Microphone Design and Assembly 2.4.1 Sensor Construction 2.4.1.1 Frequency Response 2.4.1.2 Signal to Noise Ratio Calculation 2.4.1.3 Array of Microphones 2.5 Detection of Vascular Access Stenosis Location and Severity In Vitro 2.5.1 Feature Performance 2.5.1.1 Auditory Spectral Flux (ASF) 2.5.1.2 Auditory Spectral Centroid (ASC) 2.5.2 Threshold-Based Phonoangiographic Detection of Vascular Access Stenosis 2.6 Summary of Stenosis Detection and Classification Performance 2.7 Conclusion References 3 The Temple University Hospital Digital Pathology Corpus 3.1 Introduction 3.1.1 Digital Pathology 3.1.2 Deep Learning 3.2 The TUH Digital Pathology Corpus (TUDP) 3.2.1 Computing Infrastructure 3.2.2 Image Digitization 3.2.3 Data Organization 3.2.4 Data Anonymization 3.2.5 Annotation 3.3 Deep Learning Experiments 3.3.1 Baseline System Architecture 3.3.2 Experimental Results 3.4 Summary References 4 TransientArtifactsSuppressioninTimeSeriesviaConvexAnalysis 4.1 Introduction 4.1.1 Related Work 4.2 Preliminaries 4.2.1 Difference Matrices 4.2.2 Soft-Thresholding, Total Variation, and Fused Lasso Penalty 4.2.3 The Generalized Moreau Envelope 4.3 Transient Artifacts Suppression 4.3.1 Problem Formulation 4.3.2 Optimization Algorithm 4.3.3 Parameters 4.4 The Generalized Conjoint Penalty 4.5 Transient Artifact Suppression Using the Generalized Conjoint Penalty 4.5.1 Design of Parametric Matrix B 4.5.2 Optimization Algorithm 4.6 Numerical Examples 4.6.1 Example 1 4.6.2 Example 2 4.7 Conclusion and Future Work References 5 The Hurst Exponent: A Novel Approach for Assessing Focus During Trauma Resuscitation 5.1 Introduction 5.2 Method 5.2.1 Hurst Exponent (H) 5.2.2 Experimental Protocol 5.2.3 Measure of Head Movements 5.2.4 Application of Hurst Exponent to Head Movements 5.3 Results 5.4 Discussion 5.5 Conclusion Appendix A Simplified Approach for the Estimation of the Hurst Exponent References 6 Gaussian Smoothing Filter for Improved EMG Signal Modeling 6.1 Introduction 6.2 Related Works 6.3 Problem Formulation 6.4 GSF-Based Enhanced Classification Process 6.4.1 Gaussian Smoothing Filter (GSF) 6.4.2 Filtered EMG Signals Classification 6.4.3 Support Vector Machine (SVM) 6.4.4 k-Nearest Neighbor (k-NN) 6.4.5 Naïve Bayes Classification (NBC) 6.4.6 Linear Discriminant Analysis (LDA) 6.4.7 Gaussian Mixtures Model (GMM)-Based Classifier 6.5 Experimental Validations 6.5.1 Experiment 1: Hand Gestures 6.5.2 Experiment 2: Grasping Task 6.6 Discussions 6.7 Conclusion References 7 Clustering of SCG Events Using Unsupervised Machine Learning 7.1 Introduction 7.2 Methods 7.2.1 Experimental Measurements 7.2.2 Preprocessing 7.2.2.1 Filtering 7.2.2.2 SCG Segmentation 7.2.3 Unsupervised Machine Learning 7.2.3.1 Clustering SCG Morphology 7.2.4 Dynamic Time Warping (DTW) 7.2.5 Averaging SCG Beats 7.2.5.1 DTW Barycenter Averaging (DBA) 7.2.5.2 Clustering Algorithms 7.2.5.3 k-Medoid Clustering with DTW as a Distance Measure 7.3 Results and Discussion 7.3.1 Optimum Number of Clusters 7.3.2 Purity of Clustering with Labels HLV/LLV and INS/EXP 7.3.3 Analyzing Cluster Distribution with Respiratory Phases 7.3.4 Cluster Switching 7.3.5 Relation Between Heart Rate and Clustering 7.3.6 Intra-cluster Variability 7.4 Conclusion References 8 Deep Learning Approaches for Automated Seizure Detection from Scalp Electroencephalograms 8.1 Introduction 8.1.1 Leveraging Recent Advances in Deep Learning 8.1.2 Big Data Enables Deep Learning Research 8.2 Temporal Modeling of Sequential Signals 8.2.1 A Linear Frequency Cepstral Coefficient Approach to Feature Extraction 8.2.2 Temporal and Spatial Context Modeling 8.3 Improved Spatial Modeling Using CNNs 8.3.1 Deep Two-Dimensional Convolutional Neural Networks 8.3.2 Augmenting CNNs with Deep Residual Learning 8.3.3 Unsupervised Learning 8.4 Learning Temporal Dependencies 8.4.1 Integration of Incremental Principal Component Analysis with LSTMs 8.4.2 End-to-End Sequence Labeling Using Deep Architectures 8.4.3 Temporal Event Modeling Using LSTMs 8.5 Experimentation 8.5.1 Evaluation Metrics 8.5.2 Postprocessing with Heuristics Improves Performance 8.5.3 A Comprehensive Evaluation of Hybrid Approaches 8.5.4 Optimization of Core Components 8.6 Conclusions References Correction to: The Temple University Hospital DigitalPathology Corpus Index