Biosignal and Medical Image Processing

Citation

Semmlow, J. L., & Griffel, B. (2014). Biosignal and Medical Image Processing (3rd ed.). CRC Press.

Chapter Summary

The third edition has seventeen chapters and two appendices (Numerical Integration in MATLAB; Useful MATLAB Functions), running to about 600–630 pages depending on the catalogue. Most chapters end with a summary and problems (or exercises), and the publisher says MATLAB and several of its toolboxes are used throughout. The book moves from signal basics and classical spectral analysis through filtering and time-frequency methods, to multivariate and nonlinear analysis, then to images, and ends with two chapters on classification. The publisher states that the third edition adds two new chapters on nonlinear methods for describing and classifying signals.

Chapters 1–2: Introduction; Biosignal Measurements, Noise, and Analysis

Chapter 1 covers biosignals and the measurement chain: transducers, amplifier/detector, analog signal processing and filters, analog-to-digital conversion and data banks. Chapter 2 treats biosignals themselves, noise, and signal analysis through data functions and transforms.

Chapters 3–5: Spectral Analysis: Classical Methods; Noise Reduction and Digital Filters; Modern Spectral Analysis: The Search for Narrowband Signals

Chapter 3 starts from Fourier series analysis and the power spectrum, and includes spectral averaging by Welch's method. Chapter 4 covers noise reduction by ensemble averaging, the Z-transform, and finite and infinite impulse response filters. Chapter 5 turns to parametric methods and to nonparametric eigenanalysis frequency estimation, aimed at finding narrowband signals.

Chapters 6–7: Time-Frequency Analysis; Wavelet Analysis

Chapter 6 covers the short-term Fourier transform and spectrogram, the Wigner-Ville distribution and Cohen's class distributions. Chapter 7 introduces the continuous and discrete wavelet transforms and feature detection with wavelet packets.

Chapters 8–9: Optimal and Adaptive Filters; Multivariate Analyses: Principal Component Analysis and Independent Component Analysis

Chapter 8 includes Wiener filters, adaptive signal processing and phase-sensitive detection. Chapter 9 starts with linear transformations and covers principal component analysis and independent component analysis.

Chapters 10–11: Chaos and Nonlinear Dynamics; Nonlinearity Detection: Information-Based Methods

The publisher says two chapters on nonlinear methods are new to this edition, and the contents show two such chapters here. Chapter 10 covers nonlinear systems, phase space, estimating embedding parameters, Lyapunov exponents, correlation dimension and surrogate data tests for nonlinearity. Chapter 11 covers information and regularity, the mutual information function, spectral entropy, phase-space-based entropy methods and detrended fluctuation analysis.

Chapters 12–15: Image processing

Chapter 12, Fundamentals of Image Processing: The MATLAB Image Processing Toolbox, covers image formats, display, storage, arithmetic and block-processing operations. Chapter 13, Image Processing: Filters, Transformations, and Registration, covers the two-dimensional Fourier transform, linear filtering, spatial transformations and registration. Chapter 14, Image Segmentation, covers pixel-based, continuity-based, multithresholding, morphological and edge-based methods. Chapter 15, Image Acquisition and Reconstruction, surveys imaging modalities, CT, PET and SPECT, magnetic resonance imaging and functional MRI.

Chapters 16–17: Classification I: Linear Discriminant Analysis and Support Vector Machines; Classification II: Adaptive Neural Nets

Chapter 16 covers linear discriminators, evaluating classifier performance, kernel machines, support vector machines, overfitting and cluster analysis. Chapter 17 covers training the McCulloch-Pitts neuron, the delta rule, two- and three-layer nets, training strategies and multiple classifications.

Key Concepts

Chapters 1–5: Measurement, noise and spectra

• Biosignal measurement chain: transducer, amplifier/detector, analog filtering and analog-to-digital conversion.
• Welch's method: spectral averaging for estimating power spectra.
• FIR and IIR filters: finite and infinite impulse response digital filters, with the Z-transform.
• Parametric and eigenanalysis spectral estimation: methods for narrowband signals.

Chapters 6–11: Time-frequency, adaptive and nonlinear methods

• Time-frequency distributions and wavelets: the spectrogram, Wigner-Ville distribution, and continuous and discrete wavelet transforms.
• Wiener and adaptive filters: optimal and adaptive signal processing.
• PCA and ICA: multivariate decomposition methods.
• Nonlinear dynamics: phase space, Lyapunov exponent, correlation dimension, mutual information, entropy and detrended fluctuation analysis.

Chapters 12–17: Images and classification

• Image segmentation: pixel-, continuity-, morphology- and edge-based methods.
• Image registration and reconstruction: spatial transformations, and CT, PET, SPECT and MRI reconstruction.
• Classification: linear discriminants, support vector machines and neural nets.

Critical Analysis

Strengths

• A methods toolkit written for biomedical engineers. The publisher says the book was written specifically for biomedical engineers, with additional examples using EEG, ECG, respiration and heart-rate variability, and the contents show methods beyond the classical set: wavelets, adaptive filtering, PCA/ICA, nonlinear dynamics and classifiers.
• MATLAB throughout. Concepts are implemented in MATLAB and several toolboxes, which lets readers experiment with the methods, and the publisher notes that nearly twice as many end-of-chapter problems appear in this edition.
• Designed for core courses with mixed backgrounds. The publisher's description says the book was written for a core graduate course, where a comprehensive education has to be gained in few courses, and that it also suits upper-level undergraduates; David Clifton (University of Oxford) wrote that the friendly approach suits a field whose students do not all have electrical engineering fundamentals, and Enrique Nava Baro (University of Málaga) praised the clear, simplified descriptions and the MATLAB examples. These are the reviewers quoted on the publisher's page.
• Image processing and decision-making included. The image chapters and two classification chapters extend the book from signals to images and to diagnostic decision-making.

Limitations

• Breadth trades off against depth. Seventeen chapters cover signal analysis, nonlinear methods, image processing and machine-learning classification in about six hundred pages, and the publisher itself describes the aim as covering a broad range at a "useful, working depth", so derivations and theory are necessarily selective.
• MATLAB dependence. The examples assume MATLAB and, for image processing, its Image Processing Toolbox; readers using other environments must translate the code.
• Image processing is general rather than modality-specific. The imaging modalities chapter surveys CT, PET, SPECT, MRI and fMRI within one chapter, so it is an introduction to reconstruction rather than a detailed treatment of any modality.
• Machine-learning coverage is limited to two classical topics. The classification chapters cover discriminants, SVMs and neural networks, and the contents show nothing beyond three-layer neural nets.

Where It Fits in the Course

Semmlow and Griffel is the broad methods textbook for a graduate or upper-level undergraduate biosignal course. A student who needs the electrical-engineering foundations first can start with Semmlow's Circuits, Signals, and Systems for Bioengineers, a MATLAB-based introduction built on physiological signals, or with Steven Smith's The Scientist and Engineer's Guide to Digital Signal Processing, a plain-language practical introduction to sampling, convolution, Fourier analysis and filter design. Tompkins's Biomedical Digital Signal Processing is a C-language, laboratory-oriented text from 1993 that applies signal processing to ECG, EEG and EMG signals.

For depth in particular signals, Rangayyan's Biomedical Signal Analysis is a case-study-driven treatment of ECG, EEG, EMG, heart sounds and other waveforms, developing each technique from a clinical or research problem; it complements the method-by-method organisation here. Semmlow and Griffel adds image processing and nonlinear methods to a signal-processing course.

How to Use This Book

The book suits upper-level undergraduates and graduate students in biomedical engineering, and researchers in physiology and neuroscience who process recordings. Calculus, an introductory signals and systems or circuits course, and some MATLAB are sensible preparation, although Clifton's review suggests that electrical-engineering fundamentals need not be assumed in depth.

Work through Chapters 1–4 in order, because the measurement chain, noise, spectral analysis and digital filtering underpin everything later. Chapters 5–9 can then be chosen by need: wavelets and time-frequency methods for non-stationary signals, adaptive filters for noise reduction, PCA and ICA for multichannel data. Chapters 10–11 are for readers working on nonlinear or heart-rate-variability problems. A course that includes imaging reads Chapters 12–15 as a unit, and Chapters 16–17 for diagnostic classification. Run each MATLAB example and vary its parameters, and use the problems with real recordings such as public ECG or EEG data.