No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Pattern Classification Techniques for Lung Cancer Diagnosis by an Electronic Nose
Abstract
Computational intelligence techniques can be implemented to analyze the olfactory signal as perceived by an electronic nose,
and to detect information to diagnose a multitude of human diseases. Our research suggests the use of an electronic nose to
diagnose lung cancer. An electronic nose is able to acquire and recognize the volatile organic compounds (VOCs) present in
the analyzed substance: it is composed of an array of electronic, chemical sensors, and a pattern classification module based
on computational intelligence techniques. The three main stages characterizing the basic functioning of an electronic nose
are: acquisition, preprocessing and pattern analysis. In the lung cancer detection experimentation, we analyzed 104 breath
samples of 52 subjects, 22 healthy subjects and 30 patients with primary lung cancer at different stages. In order to find
the best classification model able to discriminate between the two classes healthy and lung cancer subjects, and to reduce the dimensionality of the problem, we implemented a genetic algorithm (GA) that can find the best
combination of feature selection, feature projection and classifier algorithms to be used. In particular, for feature projection,
we considered Principal Component Analysis (PCA), Fisher Linear Discriminant Analysis (LDA) and Non Parametric Linear Discriminant
Analysis (NPLDA); classification has been performed implementing several supervised pattern classification algorithms, based
on different k-Nearest Neighbors (k-NN) approaches (classic, modified and fuzzy k-NN), on linear and quadratic discriminant
functions classifiers and on a feed-forward Artificial Neural Network (ANN). The best solution provided from the genetic algorithm
has been the projection of a subset of features into a single component using the Fisher Linear Discriminant Analysis and
a classification based on the k-Nearest Neighbors method. The observed results, all validated using cross-validation, have
been excellent achieving an average accuracy of 96.2%, an average sensitivity of 93.3% and an average specificity of 100%,
as well as very small confidence intervals. We also investigated the possibility of performing early diagnosis, building a
model able to predict a sample belonging to a subject with primary lung cancer at stage I compared to healthy subjects. Also
in this analysis results have been very satisfactory, achieving an average accuracy of 92.85%, an average sensitivity of 75.5%
and an average specificity of 97.72%. The achieved results demonstrate that the electronic nose, combined with the appropriate
computational intelligence methodologies, is a promising alternative to current lung cancer diagnostic techniques: not only
the instrument is completely non invasive, but the obtained predictive errors are lower than those achieved by present diagnostic
methods, and the cost of the analysis, both in money, time and resources, is lower. The introduction of this cutting edge
technology will lead to very important social and business effects: its low price and small dimensions allow a large scale
distribution, giving the opportunity to perform non invasive, cheap, quick, and massive early diagnosis and screening.
... E-nose has been successfully used in various applications such as: chemical engineering [19], environmental monitoring [20], cosmetic productions [21], food and beverage manufacturing [22], explosives detection [21,23], and medical diagnosis [23][24][25]. In this paper, an E-nose is designed and fabricated for odor detection in three case studies. ...
... E-nose has been successfully used in various applications such as: chemical engineering [19], environmental monitoring [20], cosmetic productions [21], food and beverage manufacturing [22], explosives detection [21,23], and medical diagnosis [23][24][25]. In this paper, an E-nose is designed and fabricated for odor detection in three case studies. ...
... Interface circuits convert sensors response to electrical signals such as voltage and current that is based on the sensors specific technology [23]. The interface circuits are designed according to the technology of Enose circuit implementation. ...
Technological progresses in the gas sensor fields provide the possibility of designing and construction of Electronic nose (E-nose) based on the Biological nose. E-nose uses specific hardware and software units; Sensor array is one of the critical units in the E-nose and its types of sensors are determined based on the application. So far, many achievements have been reported for using the E-nose in different fields of application. In this work, an E-nose for handling multi-purpose applications is proposed, and the employed hardware and pattern recognition techniques are depicted. To achieve higher recognition rate and lower power consumption, the improved binary gravitational search algorithm (IBGSA) and the K-nearest neighbor (KNN) classifier are used for automatic selecting the best combination of the sensors. The designed E-nose is tested by classifying the odors in different case studies, including moldy bread recognition in food and beverage field, herbs recognition in the medical field, and petroleum products recognition in the industrial field. Experimental results confirm the efficiency of the proposed method for E-nose realization.
... One of the difficulties in applying this method and obtaining satisfying results is to choose an accurate k number of the nearest neighbours because there are no strict or mathematical rule to determine the value of this parameter [129]. Too large k value results in a simplification problem and loose local information, on the other hand, if k is too small, the model becomes too sensitive to outliers [136]. Commonly, it is chosen after many experiments and comparing the accuracy, sensitivity, specificity metrics between different models [130]. ...
... An important element in the design of algorithms is the selection of model hyperparameters, e.g. k number of the nearest neighbors, to ensure the best generalization and prevent overfitting [136]. Model overfitting can also be reduced by using cross-validation methods such as Leave-One-One-Out Cross-Validation in the case of classic machine learning algorithms [63], [67], while using Neural Networks, popular methods of network regularization are dropout, L 2 regularization, L 1 regularization, and batch normalization [146]. ...
Currently, intensive work is underway on the development of truly noninvasive medical diagnostic systems, including respiratory analysers based on the detection of biomarkers of several diseases including diabetes. In terms of diabetes, acetone is considered as a one of the potential biomarker, although is not the single one. Therefore, the selective detection is crucial. Most often, the analysers of exhaled breath are based on the utilization of several commercially available gas sensors or on specially designed and manufactured gas sensors to obtain the highest selectivity and sensitivity to diabetes biomarkers present in the exhaled air. An important part of each system are the algorithms that are trained to detect diabetes based on data obtained from sensor matrices. The prepared review of the literature showed that there are many limitations in the development of the versatile breath analyser, such as high metabolic variability between patients, but the results obtained by researchers using the algorithms described in this paper are very promising and most of them achieve over 90% accuracy in the detection of diabetes in exhaled air. This paper summarizes the results using various measurement systems, feature extraction and feature selection methods as well as algorithms such as Support Vector Machines, k-Nearest Neighbours and various variations of Neural Networks for the detection of diabetes in patient samples and simulated artificial breath samples.
... The multi-dimensional sensor data produced by the five TGS gas sensors were given to the machine learning classification model to differentiate lung cancer patients from healthy controls, and COPD patients from healthy controls. The raw output data from the sensors were pre-processed using the baseline manipulation method before applying to the classifier [3]. From the pre-processed data, five features were extracted based on the previous studies on e-nose technologies in clinical diagnosis. ...
... Electronic noses (e-noses) are instruments that include an array of chemical sensors and can be used to recognize different odors from the analyzing samples. MOS sensor-based e-noses are nowadays used in various fields including food technology, chemical analysis, and disease diagnosis including diabetes mellitus, COPD, lung cancer and cystic fibrosis [2], [3], [5], [12], [13], [17]. Instead of identifying separate VOCs from gas samples, the MOS-based gas sensors generate a "breath signature" corresponding to different gases. ...
Purpose:
The aim of this paper was to discuss the design and development of an innovative e-nose system which can detect respiratory ailments by detecting the Volatile Organic Compounds (VOCs) in the expelled breath. In addition to nitrogen, oxygen, and carbon dioxide, the expelled breath contains several VOCs, some of which are indicative of lung-related conditions and can differentiate healthy controls from people affected with pulmonary diseases.
Methods:
This work detailed the sensor selection process, the assembly of the sensors into a sensor array, the design and implementation of the circuit, sampling methods, and an algorithm for data analysis. The clinical feasibility of the system was checked in 27 lung cancer patients, 22 chronic obstructive pulmonary disease (COPD) patients, and 39 healthy controls including smokers and non-smokers.
Results:
The classification model developed using the support vector machine (SVM) was able to provide accuracy, sensitivity, and specificity of 88.79, 89.58 and 88.23%, respectively for lung cancer, and 78.70, 72.50 and 82.35%, respectively for COPD.
Conclusions:
The sensor array system developed with TGS gas sensors was non-invasive, low cost, and gave a rapid response. It has been demonstrated that the VOC profiles of patients with pulmonary diseases and healthy controls are different, hence, the e-nose system can be used as a potential diagnostic device for patients with lung diseases.
... The features include the transient signals, steady-state signals, the parameters of various response curve-fitting models, and the coefficients of different transforms. The features extracted from the sensor response curves are as follows [21]. ...
Breath analysis has become a promising tool for the detection of pulmonary diseases in recent years. This paper describes the fabrication of an artificial intelligence (AI) based e-nose system for discriminating lung cancer patients from healthy controls. Breath volatile organic compounds (VOC) can be easily analyzed using an electronic nose (e-nose) made of various gas sensor arrays. Metal oxide semiconductor (MOS)-based e-noses are getting popular in the VOC analysis of exhaled breath from humans. Here we have developed the e-nose system with a sensor array system of five MOS gas sensors and incorporated the controller and machine learning algorithms to process the data. The sensor array was designed with MOS sensors developed by Figaro USA and the data acquisition was carried out with the help of the Arduino Uno developer board. 40 healthy control samples and 24 lung cancer patient samples were analyzed using the developed e-nose system. The data analysis was done by two supervised classification algorithms random forest and logistic regression. Among this, random forest with 5-fold cross-validation gave better results with 85.38 % classification accuracy and 0.87 of AUC. This system can be further extended to the diagnosis of various other pulmonary diseases.
... However, if this method implemented in real systems with long time data collection, the drift should be taken into concern. Besides, concerning the time constraint for the lung cancer volatile detection application, it is inconvenient for the patient for several minutes during the measurement of the steady state features [13]. Since, the transient response also able to give high performance as the steady state response, it is confirmed that the features from transient response contains useful information related to dynamics of phenomenon and with the help of LDA analysis, it can be used as a substitution of the steady state method to accelerate the volatile detection process. ...
In this experiment, three different cell cultures (A549, WI38VA13 and MCF7) and blank medium (without cells) as a control were used. The electronic nose (E-Nose) was used to sniff the headspace of cultured cells and the data were recorded. After data pre-processing, two different features were extracted by taking into consideration of both steady state and the transient information. The extracted data are then being processed by multivariate analysis, Linear Discriminant Analysis (LDA) to provide visualization of the clustering vector information in multi-sensor space. The Probabilistic Neural Network (PNN) classifier was used to test the performance of the E-Nose on determining the volatile organic compounds (VOCs) of lungcancercell line. The LDA data projection was able to differentiate between the lungcancercell samples and other samples (breast cancer, normal cell and blank medium) effectively. The features extracted from the steady state response reached 100% of classification rate while the transient response with the aid of LDA dimension reduction methods produced 100% classification performance using PNN classifier with a spread value of 0.1. The results also show that E-Nose application is a promising technique to be applied to real patients in further work and the aid of Multivariate Analysis; it is able to be the alternative to the current lungcancer diagnostic methods.
... For example, if K is too large, then the problem will be simplified and the local information will be lost. On the other hand, when K is small, the density estimation will be too much sensitive to the outliers [10]. In the KNN classifier with the k=1 and with infinite number of training samples, the minimum error is never higher than twice the of the Bayesian error [16]. ...
For a long time, human odors and vapors have been known for their diagnostic power. Therefore, the analysis of the metabolic composition of human breath and odors creates the opportunity for a non-invasive tool for clinical diagnostics. Innovative analytical technologies to capture the metabolic profile of a patient's breath are available, such as, for instance, the ion mobility spectrometry coupled to a multicapilary collumn. However, we are lacking automated systems to process, analyse and evaluate large clinical studies of the human exhaled air.
To fill this gap, a number of computational challenges need to be addressed. For instance, breath studies generate large amounts of heterogeneous data that requires automated preprocessing, peak-detection and identification as a basis for a sophisticated follow up analysis. In addition, generalizable statistical evaluation frameworks for the detection of breath biomarker profiles that are robust enough to be employed in routine clinical practice are necessary. In particular since breath metabolomics is susceptible to specific confounding factors and background noise, similar to other clinical diagnostics technologies. Moreover, spesific manifestations of disease stages and progression, may largely influence the breathomics profiles.
To this end, this thesis will address these challenges to move towards more automatization and generalization in clinical breath research. In particular I present methods to support the search for biomarker profiles that enable a non-invasive detection of diseases, treatment optimization and prognosis to provide a new powerful tool for precision medicine.
http://scidok.sulb.uni-saarland.de/volltexte/2016/6587/
We define breathomics as the metabolomics study of exhaled air. It is a strongly emerging metabolomics research field that mainly focuses on health-related volatile organic compounds (VOCs). Since the amount of these compounds varies with health status, breathomics holds great promise to deliver non-invasive diagnostic tools. Thus, the main aim of breathomics is to find patterns of VOCs related to abnormal (for instance inflammatory) metabolic processes occurring in the human body. Recently, analytical methods for measuring VOCs in exhaled air with high resolution and high throughput have been extensively developed. Yet, the application of machine learning methods for fingerprinting VOC profiles in the breathomics is still in its infancy. Therefore, in this paper, we describe the current state of the art in data pre-processing and multivariate analysis of breathomics data. We start with the detailed pre-processing pipelines for breathomics data obtained from gas-chromatography mass spectrometry and an ion-mobility spectrometer coupled to multi-capillary columns. The outcome of data pre-processing is a matrix containing the relative abundances of a set of VOCs for a group of patients under different conditions (e.g. disease stage, treatment). Independently of the utilized analytical method, the most important question, 'which VOCs are discriminatory?', remains the same. Answers can be given by several modern machine learning techniques (multivariate statistics) and, therefore, are the focus of this paper. We demonstrate the advantages as well the drawbacks of such techniques. We aim to help the community to understand how to profit from a particular method. In parallel, we hope to make the community aware of the existing data fusion methods, as yet unresearched in breathomics.
Introduction:
The early determination of serious pathologies has so far been an important issue in both the medical and social fields. The search for an instrument able to detect cancers has led to the consideration of the usage of chemicals of the human body, which carry, through its volatile compounds, information coming from or related to defined pathologies.
Areas covered:
The electronic nose (EN) seems to represent a good solution for the detection of cancers of different types. Recent results showed the utility of an EN to smell chemicals related to lung, melanoma, prostatic, breast and pancreatic cancers. The results obtainable from ENs are chemical images and, as it will be shown in this paper, the probability of cancer recognition is rather high. Main results obtained at international level and by the authors of this paper will be commented upon.
Expert opinion:
A personal opinion is given trying to foresee future developments of the olfaction strategy. To this purpose, two main aspects are considered: looking for better overall stability of the EN and for a new use of ENs in detecting alterations between blood and pathology components.
We present a method to diagnose lung cancer by the analysis of breath using an electronic nose. This device can react to a gas substance by providing signals that can be analyzed to classify the input. It is composed of a sensor array (6 MOS sensors, in our case) and a pattern classification process based on machine learning techniques. During the first phase of our research, we have evaluated the possibility and accuracy of lung cancer diagnosis by classifying the olfactory signal associated to exhalations of subjects. The second part of the research, still in progress, is aimed at assessing the possibility of discriminating also the different types and stages of the disease. At the end of the first phase, results have been very satisfactory and promising: we achieved an average accuracy of 92.6%, sensitivity of 95.3% and specificity of 90.5%. In particular we analyzed the breath of 101 individuals, of which 58 control subjects, and 43 suffer from different types of lung cancer (primary and not) at different stages. In order to find the components able to discriminate between the two classes 'healthy' and 'sick' at best, and to reduce the dimensionality of the problem, we have extracted the most significant features and projected them into a lower dimensional space using Non Parametric Linear Discriminant Analysis. Finally, we have used these features as input to several supervised pattern classification algorithms, based on different k-nearest neighbors (k-NN) approaches (classic, modified and Fuzzy k-NN), linear and quadratic discriminant classifiers and on a feed-forward artificial neural network (ANN). The observed results have all been validated using cross-validation. These results pushed us to begin the second phase of the project to investigate the possibility of early lung cancer diagnosis: we are involving a larger number of subjects, partioned in different classes according to the type and stage of the disease. The research demonstrates that the electronic nose is a promising alternative to current lung cancer diagnostic techniques: the obtained predictive errors are lower than those achieved by present diagnostic methods, and the cost of the analysis, both in money, time and resources, is lower. The introduction of this technology will lead to very important social and business effects: its low price and small dimensions allow a large scale distribution, giving the opportunity to perform non invasive, cheap, quick, and massive early diagnosis and screening.
Using a specially developed breath collection technique and computer-assisted gas chromatography/mass spectrometry (GC/MS), we have identified in the exhaled air of lung cancer patients several volatile organic compounds that appear to be associated with the disease. The GC/MS profiles of 12 samples from lung cancer patients and 17 control samples were analyzed by using general computerized statistical procedures to distinguish lung cancer patients from controls. The selected volatile compounds had sufficient diagnostic power in the GC/MS profiles to allow almost complete differentiation between the two groups in a limited patient population.
Witnessing the swift advances in the electronic means of seeing
and hearing, scientists and engineers scent a market for systems
mimicking the human nose. Already commercial systems from several
companies are targeting applications, present and potential, that range
from quality assurance of food and drugs to medical diagnosis,
environmental monitoring, safety and security and military use. Here,
the authors outline the major transducer technologies-in one sense, the
key component of an electronic nose
Study objectives
To evaluate volatile organic compounds (VOCs) in the breath as tumor markers in lung cancer. Alkanes and monomethylated alkanes are oxidative stress products that are excreted in the breath, the catabolism of which may be accelerated by polymorphic cytochrome p450-mixed oxidase enzymes that are induced in patients with lung cancer.
Design
Combined case-control and cross-sectional study.
Setting
Five academic pulmonary medicine services in the United States and the United Kingdom.
Patients and participants
One hundred seventy-eight bronchoscopy patients and 41 healthy volunteers.
Intervention
Breath samples were analyzed by gas chromatography and mass spectroscopy to determine alveolar gradients (ie, the abundance in breath minus the abundance in room air) of C4-C20 alkanes and monomethylated alkanes.
Measurements
Patients with primary lung cancer (PLC) were compared to healthy volunteers, and a predictive model was constructed using forward stepwise discriminant analysis of the alveolar gradients. This model was cross-validated with a leave-one-out jackknife technique and was tested in two additional groups of patients who had not been used to develop the model (ie, bronchoscopy patients in whom cancer was not detected, and patients with metastatic lung cancer [MLC]).
Results
Eighty-seven of 178 patients had lung cancer (PLC, 67 patients; MLC, 15 patients; undetermined, 5 patients). A predictive model employing nine VOCs identified PLC with a sensitivity of 89.6% (60 of 67 patients) and a specificity of 82.9% (34 of 41 patients). On cross-validation, the sensitivity was 85.1% (57 of 67 patients) and the specificity was 80.5% (33 of 41 patients). The stratification of patients by tobacco smoking status, histologic type of cancer, and TNM stage of cancer revealed no marked effects. In the two additional tests, the model predicted MLC with a sensitivity of 66.7% (10 of 15 patients), and it classified the cancer-negative bronchoscopy patients with a specificity of 37.4% (34 of 91 patients).
Conclusions
Compared to healthy volunteers, patients with PLC had abnormal breath test findings that were consistent with the accelerated catabolism of alkanes and monomethylated alkanes. A predictive model employing nine of these VOCs exhibited sufficient sensitivity and specificity to be considered as a screen for lung cancer in a high-risk population such as adult smokers.
Considerable interest has recently arisen in the use of arrays of gas sensors together with an associated pattern-recognition technique to identfy vapours and odours. The performance of the pattern-recognition technique depends upon the choice of parametric expression used to define the array output. At present, there is no generally agreed choice of this parameter for either individual sensors or arrays of sensors. In this paper, we have initially performed a parametric study on experimental data gathered from the response of an array of twelve tin oxide gas sensors to five alcohols and three beers. Five parametric expressions of sensor response are used to characterize the array output, namely, fractional conductance change, relative conductance, log of conductance change and normalized versions of the last two expressions. Secondly, we have applied the technique of artificial neural networks (ANNs) to our preprocessed data. The Rumelhart back-propagation technique is used to train all networks. We find that nearly all of our ANNs can correctly identify all the alcohols using our array of twelve tin oxide sensors and so we use the total sum of squared network errors to determine their relative performance. It is found that the lowest network error occurs for the response parameter defined as the fractional change in conductance, with a value of 1.3 × 10−4, which is almost half that for the relative conductance. The normalized procedure is also found to improve network performance and so is worthwhile. The optimal network for our data-set is found to contain a single hidden layer of seven elements with a learning rate of 1.0 and momentum term of 0.7, rather than the values of 0.9 and 0.6 recommended by Rumelhart and McClelland, respectively. For this network, the largest output error is less than 0.1. We find that this network outperforms principal-component and cluster analyses (discussed in Part 1) by identifying similar beer odours and offers considerable benefit in its ability to cope with non-linear and highly correlated data.
Classification of objects is an important area of research and application in a variety of fields. In the presence of full knowledge of the underlying probabilities, Bayes decision theory gives optimal error rates. In those cases where this information is not present, many algorithms make use of distance or similarity among samples as a means of classification. The K-nearest neighbor decision rule has often been used in these pattern recognition problems. One of the difficulties that arises when utilizing this technique is that each of the labeled samples is given equal importance in deciding the class memberships of the pattern to be classified, regardless of their 'typicalness'. The theory of fuzzy sets is introduced into the K-nearest neighbor technique to develop a fuzzy version of the algorithm. Three methods of assigning fuzzy memberships to the labeled samples are proposed, and experimental results and comparisons to the crisp version are presented.
"Electronic noses" are instruments which mimic the sense of smell.
Consisting of olfactory sensors and a suitable signal processing unit,
they are able to detect and distinguish odors precisely and at low cost.
This makes them very useful for a remarkable variety of applications in
the food and pharmaceutical industry, in environmental control or
clinical diagnostics and more. The scope covers biological and technical
fundamentals and up-to-date research. Contributions by renowned
international scientists as well as application-oriented news from
successful "e-nose" manufacturers give a well-rounded account of the
topic, and this coverage from R&D to applications makes this book a
must-have read for e-nose researchers, designers and users alike.
The measurement and estimation of human-related senses has become an established technique in sensor research, as well as in the practical design of measurement and control systems. The electronic nose concept is widely used as an analytical tool in industry today. The commercialization of the electronic nose began in 1993 as the concept became widely accepted as an effective instrument for detection and estimation of olfaction. The book describes the general set-up of an electronic nose: it consists of an array of chemical sensors; an air flow system, which switches the reference air and the tested air; a signal analysis technique; and a presentation unit. The main sensor principles presented in the book are also the most frequently used techniques for gas sensors. These are based on two main types of gas sensor: metal-oxide semiconductors and conducting polymer resistive materials. An overview of other gas sensor principles is also given, for example, gas sensors based on the effect of sorbed molecules on the propagation of acoustic waves; field effect semiconductors; electrochemical oxidation and reduction principles; and a catalytic gas sensor. Pellistors are described as well as fibre-optic gas sensors. To increase the complexity of the odour system, an array of mixed sensing principles is often designed, consisting of different types of sensor, in order to create differences in operating temperatures, flow conditions and sensor response times. The analysis technique used, in most cases, is a supervised artificial neural network used in a relative based measurement approach, although other techniques are also mentioned in the book.
The book describes a survey of both chemical properties of odorous molecules as well as biological olfaction from the perspective of human perception. However, many animals have more sophisticated systems, and it would have been interesting to make comparisons.
The applications and three case studies show the direction of the research and development over the last decade. However, the presentation of most of the results as principal component plots indicates that major work is needed to achieve a more human-friendly interface. The next decade should see the development of electronic nose systems into a variety of applications to increase the quality of life as well as for monitoring environmental information. This means that artificial human-related sensor systems could become everyday tools for estimation of your own personal condition as well as that of the environment.
In general, the book presents the state-of-the-art in electronic noses. It can also be considered as an introduction to the subject of electronic noses. This opinion is supported by a sufficiently complete set of references and a good introduction to the chemical, biological and physical background of electronic noses. The book is characterized by a methodical and thorough treatment of the subject matter. The chapters are logically related, and each has its own introduction and bibliography. This approach makes it easy for any reader with only a basic knowledge of related subjects.
However, for researchers, the material and examples feel a little old, since the last three or four years have seen an enormous amount of activity in the field of human-related artificial sensor systems, where the electronic nose is often integrated into more complex systems. By fusing sensor data into a hybrid sensor, or by more complex signal analysis combining data from other sensor modalities such as electronic tongues, we aim to make inferences that may be impossible from a single artificial sensor. Furthermore, a whole electronic head integrates information from all five human-related artificial senses.
However, it remains true that, in a particular area strongly relevant to human-related quality estimation, this book is a valuable and accessible guide and reference. I strongly recommend it as a handbook for users of commercial electronic nose instruments and, particularly for its references to related works.
Peter Wide
Quantitative analysis of gases, by means of semiconductor sensor arrays and pattern-recognition techniques such as artificial neural networks, has been the goal of a great deal of work over the last few years. However, the lack of selectivity, repeatability and drifts of the sensors, have limited the applications of these systems to qualitative or semi-quantitative gas analysis. While the steady-state response of the sensors is usually the signal to be processed in such analysis systems, our method consists of processing both, transient and steady-state information. The sensor transient behaviour is characterised through the measure of its conductance rise time (Tr), when there is a step change in the gas concentration. Tr is characteristic of each gas/sensor pair, concentration-independent and shows higher repeatability than the steady state measurements. An array of four thick-film tin oxide gas sensors and pattern-recognition techniques are used to discriminate and quantify among ethanol, toluene and o-xylene [concentration range: 25, 50 and 100 ppm]. A principal component analysis is carried out to show qualitatively that selectivity improves when the sensor behaviour is dynamically characterised. The steady-state and transient conductance of the array components are processed with artificial neural networks. In a first stage, a feed-forward back-propagation-trained ANN discriminates among the studied compounds. Afterwards, three separate ANN (one for each vapour) are used to quantify the previously identified compound. Processing data from the dynamic characterisation of the sensor array, considerably improves its identification performance, rising the discrimination success rate from a 66% when only steady-state signals are used up to 100%.
A fuzzy set is a class of objects with a continuum of grades of membership. Such a set is characterized by a membership (characteristic) function which assigns to each object a grade of membership ranging between zero and one. The notions of inclusion, union, intersection, complement, relation, convexity, etc., are extended to such sets, and various properties of these notions in the context of fuzzy sets are established. In particular, a separation theorem for convex fuzzy sets is proved without requiring that the fuzzy sets be disjoint.
This book provides the first comprehensive treatment of feed-forward neural networks from the perspective of statistical pattern recognition. After introducing the basic concepts of pattern recognition, the book describes techniques for modelling probability density functions, and discusses the properties and relative merits of the multi-layer perceptron and radial basis function network models. It also motivates the use of various forms of error functions, and reviews the principal algorithms for error function minimization. As well as providing a detailed discussion of learning and generalization in neural networks, the book also covers the important topics of data processing, feature extraction, and prior knowledge. The book concludes with an extensive treatment of Bayesian techniques and their applications to neural networks.
A nonparametric method of discriminant analysis is proposed. It is based on nonparametric extensions of commonly used scatter matrices. Two advantages result from the use of the proposed nonparametric scatter matrices. First, they are generally of full rank. This provides the ability to specify the number of extracted features desired. This is in contrast to parametric discriminant analysis, which for an L class problem typically can determine at most L 1 features. Second, the nonparametric nature of the scatter matrices allows the procedure to work well even for non-Gaussian data sets. Using the same basic framework, a procedure is proposed to test the structural similarity of two distributions. The procedure works in high-dimensional space. It specifies a linear decomposition of the original data space in which a relative indication of dissimilarity along each new basis vector is provided. The nonparametric scatter matrices are also used to derive a clustering procedure, which is recognized as a k-nearest neighbor version of the nonparametric valley seeking algorithm. The form which results provides a unified view of the parametric nearest mean reclassification algorithm and the nonparametric valley seeking algorithm.
Part I of this paper describes a model for the parallel development and adult coding of neural feature detectors. It shows how any set of arbitrary spatial patterns can be recoded, or transformed, into any other spatial patterns (universal recoding), if there are sufficiently many cells in the network's cortex. This code is, however, unstable through time if arbitrarily many patterns can perturb a fixed number of cortical cells. This paper shows how to stabilize the code in the general case using feedback between cellular sites. A biochemically defined critical period is not necessary to stabilize the code, nor is it sufficient to ensure useful coding properties.We ask how short term memory can be reset in response to temporal sequences of spatial patterns. This leads to a context-dependent code in which no feature detector need uniquely characterize an input pattern; yet unique classification by the pattern of activity across feature detectors is possible. This property uses learned expectation mechanisms whereby unexpected patterns are temporarily suppressed and/or activate nonspecific arousal. The simplest case describes reciprocal interactions via trainable synaptic pathways (long term memory traces) between two recurrent on-center off-surround networks undergoing mass action (shunting) interactions. This unit can establish an adaptive resonance, or reverberation, between two regions if their coded patterns match, and can suppress the reverberation if their patterns do not match. This concept yields a model of olfactory coding within the olfactory bulb and prepyriform cortex. The resonance idea also includes the establishment of reverberation between conditioned reinforcers and generators of contingent negative variation if presently avialable sensory cues are compatible with the network's drive requirements at that time; and a search and lock mechanism whereby the disparity between two patterns can be minimized and the minimal disparity images locked into position. Stabilizing the code uses attentional mechanisms, in particular nonspecific arousal as a tuning and search device. We suggest that arousal is gated by a chemical transmitter system—for example, norepinephrine—whose relative states of accumulation at antagonistic pairs of on-cells and off-cells through time can shift the spatial pattern of STM activity across a field of feature detectors. For example, a sudden arousal increment in response to an un-expected pattern can reverse, or rebound, these relative activities, thereby suppressing incorrectly classified populations. The rebound mechanism has formal properties analogous to negative afterimages and spatial frequency adaptation.
This paper analyses a model for the parallel development and adult coding of neural feature detectors. The model was introduced in Grossberg (1976). We show how experience can retune feature detectors to respond to a prescribed convex set of spatial patterns. In particular, the detectors automatically respond to average features chosen from the set even if the average features have never been experienced. Using this procedure, any set of arbitrary spatial patterns can be recoded, or transformed, into any other spatial patterns (universal recoding), if there are sufficiently many cells in the network's cortex. The network is built from short term memory (STM) and long term memory (LTM) mechanisms, including mechanisms of adaptation, filtering, contrast enhancement, tuning, and nonspecific arousal. These mechanisms capture some experimental properties of plasticity in the kitten visual cortex. The model also suggests a classification of adult feature detector properties in terms of a small number of functional principles. In particular, experiments on retinal dynamics, including amarcrine cell function, are suggested.
Handbook of machine olfaction. Electronic nose technology
- T C Pearce
- H T Nagle
- S S Shiffman
- J W Gardner
- T.C. Pearce