Artificial Neural Networks in Biomedicine

by Lisboa, P. J. G.; Ifeachor, Emmanuel C.; Szczepaniak, Piotr S.

• ISBN13: 9781852330057
• ISBN10: 1852330058
• Format: Paperback
• Copyright: 1999-08-01
• Publisher: Springer-Nature New York Inc

Summary

This volume provides a state-of-the-art survey of artificial neural network applications in biomedical diagnosis, laboratory data analysis and related practical areas. It looks at biomedical applications which involve customising neural network technology to resolve specific difficulties with data processing, and deals with applications relating to particular aspects of clinical practice and laboratory or medically-related analysis. Each chapter is self-contained with regard to the technology used, covering important technical points and implementation issues like the design of user interfaces and hardware/software platforms. Artificial Neural Networks in Biomedicine will be of interest to computer scientists and neural network practitioners who want to extend their knowledge of issues relevant to biomedical applications, developers of clinical computer systems, and medical researchers looking for new methods and computational tools.

Table of Contents

Introduction	1	(8)
TUTORIAL AND REVIEW	9	(40)
The Bayesian Paradigm: Second Generation Neural Computing	11	(14)
Introduction	11	(1)
Theory	12	(8)
Bayesian Learning	13	(2)
The Evidence Framework	15	(1)
Error bars	16	(1)
Moderated outputs	16	(1)
Regularisation	17	(1)
Committees	18	(2)
Example Results	20	(1)
Conclusion	21	(4)
The Role of the Artificial Neural Network in the Characterisation of Complex Systems and the Prediction of Disease	25	(14)
Introduction	26	(2)
Diagnosis of Disease	28	(2)
Outcome Prediction	30	(1)
Conclusion	31	(8)
Genetic Evolution of Neural Network Architectures	39	(10)
Introduction	39	(1)
Stability: The `Bias/Variance Problem'	40	(1)
Genetic Algorithms and Artificial Neural Networks	41	(5)
Description of a General Method for Evolving ANN Architecture (EANN)	42	(1)
Prediction of Depression After Mania	43	(1)
EANN and the Agreement/Transparency Choice	43	(2)
ANN and the Stability/Specialisation Choice	45	(1)
Conclusion	46	(3)
COMPUTER AIDED DIAGNOSIS	49	(102)
The Application of PAPNET to Diagnostic Cytology	51	(18)
Introduction	51	(1)
First Efforts at Automation in Cytology	52	(1)
Neural Networks	53	(1)
The PAPNET System®	53	(10)
Components of the PAPNET System	54	(5)
Technical factors affecting the performance of the machine	59	(1)
Performance of the PAPNET System	59	(1)
Cervicovaginal smears	59	(2)
Application of the PAPNET System to Smears of Sputum	61	(1)
Application of the PAPNET System to Smears of Urinary Sediment	61	(1)
Application of the PAPNET System to Oesophageal Smears	62	(1)
Comment	63	(6)
ProstAsure Index -- A Serum-Based Neural Network-Derived Composite Index for Early Detection of Prostate Cancer	69	(12)
Introduction	69	(1)
Clinical Background of Prostate Cancer and Derivation of the ProstAsure Index Algorithm	70	(2)
Validation of PI with Independent Clinical Data	72	(1)
Issues in Developing PI	73	(3)
Conclusion	76	(5)
Neurometric Assessment of Adequacy of Intraoperative Anaesthetic	81	(12)
Intraoperative Awareness	81	(1)
Measuring Sensory Perception	82	(1)
Clinical Data	82	(1)
Results	83	(3)
Implementation	86	(2)
Clinical Deployment	88	(1)
Healthcare Benefit	89	(1)
Additional Studies	89	(4)
Classifying Spinal Measurements Using a Radial Basis Function Network	93	(12)
Introduction	93	(1)
Data	94	(2)
The Spines	94	(1)
The Measurements	94	(1)
Preprocessing the Data	95	(1)
Radial Basis Functions and Networks	96	(1)
Matrix Notation	97	(1)
Training RBF Networks	98	(4)
The Unsupervised Learning Stage	98	(1)
The Supervised Learning Stage	98	(1)
Regularisation as an aid to avoid over-fitting	98	(1)
Calculating the regularisation coefficients and the weights	99	(2)
Forward subset selection of RBFs	101	(1)
Input feature selection	102	(1)
Results	102	(1)
Conclusion	103	(2)
GEORGIA: An Overview	105	(12)
Introduction	106	(1)
The Medical Decision Support System	107	(2)
Learning Pattern Generation	109	(1)
Software and Hardware Implementation	110	(1)
Re-Training and Re-Configuring the MDSS	111	(1)
Introducing GEORGIA's Man-to-Computer Interface	111	(3)
Conclusion	114	(3)
Patient Monitoring Using an Artificial Neural Network	117	(12)
Overview of the Medical Context	117	(1)
Basic Statistical Appraisal of Vital Function Data	118	(2)
Neural Network Details	120	(3)
Default Training	121	(2)
Implementation	123	(1)
Clinical Trials	124	(1)
Clinical Practice	125	(4)
Benchmark of Approaches to Sequential Diagnosis	129	(12)
Introduction	129	(1)
Preliminaries	130	(2)
Methods	132	(5)
The Probabilistic Algorithm	132	(1)
The diagnostic algorithm for first order markov chains- the Markov I algorithm	132	(1)
The diagnostic algorithm for second order markov chains-the Markov II algorithm	133	(2)
The Fuzzy Methods	135	(1)
The algorithm without context - fuzzy 0	135	(1)
The algorithm with first-order context - fuzzy 1A	135	(1)
The reduced algorithm with first-order context - fuzzy 1B	135	(1)
The algorithm with second-order context - fuzzy 2A	135	(1)
The reduced algorithm with second-order context - fuzzy 2B	136	(1)
The Neural Network Approach	136	(1)
A Practical Example - Comparative Analysis of Methods	137	(1)
Conclusion	138	(3)
Application of Neural Networks in the Diagnosis of Pathological Speech	141	(10)
Introduction	141	(1)
The Research Material and the Problems Considered	142	(3)
Dental Prosthetics	142	(1)
Maxillofacial Surgery	143	(1)
Orthodontics	144	(1)
Laryngology	144	(1)
The Signal Parameterisation	145	(2)
The Application of the Neural Networks and the Results	147	(2)
Conclusion	149	(2)
SIGNAL PROCESSING	151	(60)
Independent Components Analysis	153	(16)
Introduction	153	(1)
Theory	153	(5)
The Decorrelating Manifold	155	(1)
The Choice of Non-Linearity	156	(2)
Model-Order Estimation	158	(1)
Non-Stationary ICA	158	(5)
Illustration	161	(2)
Applications	163	(3)
Source Separation	164	(1)
Source Number and Estimation	164	(2)
Conclusion	166	(3)
Rest EEG Hidden Dynamics as a Discriminant for Brain Tumour Classification	169	(12)
Introduction	170	(1)
Characterising Hidden Dynamics	171	(3)
The Clinical Study	174	(2)
The Minimum Markov Order	176	(3)
Conclusion	179	(2)
Artifical Neural Network Control on Functional Electrical Stimulation Assisted Gait for Persons with Spinal Cord Injury	181	(14)
Introduction	182	(1)
Methods	183	(4)
Results	187	(4)
Discussion	191	(4)
The Application of Neural Networks to Interpret Evoked Potential Waveforms	195	(16)
Introduction	195	(1)
The Medical Conditions Studied	196	(1)
The Evoked Potentials	196	(1)
The Relationship Between the CNV and the Medical Conditions	197	(1)
Experimental Procedures	198	(1)
Data Pre-Processing	198	(1)
Feature Extraction	199	(1)
Normalisation	200	(1)
The Artificial Neural Networks	200	(6)
The Simplified Fuzzy ARTMAP	200	(4)
The Probabilistic Simplified Fuzzy ARTMAP	204	(1)
ANN Training and Accuracy	205	(1)
Small numbers of training vectors	205	(1)
Simplified fuzzy ARTMAP	205	(1)
Committees of ANNs	206	(1)
Validation Issues	206	(1)
Technical Aspects of Validation	206	(1)
Clinical Aspects of Validation	206	(1)
Results	207	(1)
Implementation Considerations	207	(1)
Future Developments	208	(3)
IMAGE PROCESSING	211	(72)
Intelligent Decision Support Systems in the Cytodiagnosis of Breast Carcinoma	213	(20)
Introduction	213	(2)
Previous Work on Decision Support in this Domain	215	(1)
The Data Set in this Study	215	(11)
Study Population	215	(1)
Input Variables	216	(1)
Partitioning of the Data	217	(1)
Human Performance	217	(1)
Logistic Regression	218	(1)
Data Derived Decision Tree	219	(1)
Multi-Layer Perceptron Neural Networks	220	(2)
Adaptive Resonance Theory Mapping (ARTMAP) Neural Networks	222	(1)
Potential Advantages of ARTMAP	222	(1)
ARTMAP Architecture and Methodology	222	(3)
Results from the Cascaded System	225	(1)
Symbolic Rule Extraction	225	(1)
Assessment of the Different Decision Support Systems	226	(7)
A Neural-Based System for the Automatic Classification and Follow-Up of Diabetic Retinopathies	233	(16)
Introduction	233	(2)
The DRA System	235	(2)
Hybrid Module	237	(2)
Committee Algorithms	239	(6)
New Selection Algorithms	240	(1)
Greedy selection	241	(1)
Pseudo-exhaustive selection	241	(1)
Sequential Cooperation	242	(1)
Experimental Results	243	(2)
Related Work	245	(1)
Validation of the DRA System	245	(1)
Conclusion	246	(3)
Classification of Chromosomes: A Comparative Study of Neural Network and Statistical Approaches	249	(18)
Introduction	249	(3)
Chromosome Analysis and its Applications	249	(1)
Chromosome Classification	250	(1)
Experimental Data	251	(1)
The Neural Network Classifier	252	(3)
Representation of Chromosome Features	252	(1)
Network Topology and Training	253	(1)
Incorporating Non-Banding Features	254	(1)
Classification Performance	255	(3)
Classification Experiments	255	(1)
Comparison with Statistical Classifiers	256	(1)
The Influence of Training-Set Size	256	(2)
The Use of Context in Classification	258	(3)
The Karyotyping Constraint	258	(1)
Applying the Constraint by a Network	259	(1)
Results of Applying the Context Network	260	(1)
Conclusion and Discussion	261	(6)
Comparison with Statistical Classifiers	261	(1)
Training Set Size and Application of Context	262	(1)
Biological Context	263	(4)
The Importance of Features and Primitives for Multi-dimensional/Multi-channel Image Processing	267	(16)
Introduction	267	(2)
The Image Data Level	269	(1)
From Image Data to Symbolic Primitives	269	(1)
Region Segmentation Quality and Training Phase	270	(1)
Validation of Image Segmentation	271	(2)
Segmentation Complexity and Quantitative Error Evaluation	273	(2)
Feature Description	275	(1)
Feature Selection	276	(2)
A Preliminary Overview of Application Results	278	(3)
Conclusion	281	(2)
Index	283