REVIEW OF RELATED LITERATURE
Artificial Neural Networks (ANNs) has already established itself both in disease detection and treatment (Ochoa, Chana and Stud, 2001; Fraser, 2000; Almeida, 2002; Agatonovic-Kustrin and Beresford, 2000) whether in the field of cancer (Bostwick and Burke, 2001; Grumett, Snow and Kerr, 2003; Khan, et. al. 2001; Seker et. al., 2002; Jerez-Aragones, Gomez-Ruiz, Ramos-Jimenez, Munoz-Perez and Alba-Conejo, 2003; Hung, Shanker and Hu, 2002; Abbass, 2002; Wallace, Bamber, Crawford, Ott, and Mortimer, 2000; Babaian and Zhang, 2001); neurobiology (Basheer and Hajmeer, 2000; Maas and Bishop, 2001; Gluck and Myers, 2001; Agatonovic-Kustrin and Beresford, 2000; Mihailidis, Fernie and Barbenel, 2001; Van Hoey et. al., 2000); genetic analysis (Amendolia, et al., 2002); oncological prognosis (Seker, Odetayo, Petrovic and Naguib 2003); cardiovascular treatment (Folland, Hines, Boilot and Morgan, 2002; Newey and Nassiri, 2002); chronic fatigue (Linder, Dinser, Wagner, Krueger and Hoffmann, 2002); pulmonary embolism (Eng, 2002); ICU treatment (Parmanto, Deneault, and Denault, 2001; Frize, Ennett, Stevenson, and Trigg, 2001; Tong, Frize, and Walker, 2002; Van Hoey et al. 2000); and in pharmaceuticals (Agatonovic-Kustrin and Beresford, 2000; Hajmeer and Basheer, 2002; Shang, Lin, and Goetz, 2000).
This review of the literature shall discuss the uses of Artificial Neural Networks in medicine; the accuracy in detecting the disease and in finding ways to better improve the treatment. Particular in this study is the use of ANN in cancer detection and treatment, neurobiology, design of medicines or pharmaceutical and ICU treatment. Moreover, oncological prognosis, chronic fatigue, pulmonary embolism and fluoride detection shall also be discussed. The contribution of ANN rests on its ability to predict more accurately thus, this review shall also discuss a comparative result of the ANN model vis a vis the other linear models.
Artificial Neural Networks
Neural network based methodologies are used for image interpretation in many hospitals around the world. The use of this AI based technology increasingly provides opportunities to facilitate and enhance the work of medical experts and ultimately to improve the efficiency and quality of medical care.
The main features in medical diagnosis and prediction using artificial intelligence techniques will make the consultation to be more interactive (Ishak and Siraj, 2002). As clinical decision making inherently requires reasoning under uncertainty, expert systems and fuzzy logic will be suitable techniques for dealing with partial evidence and with uncertainty regarding the effects of proposed interventions. For the prediction tasks, Neural Networks have been proven to produce better results compared to other techniques (such as statistics) (Ishak and Siraj, 2002; Errejon, et. al., 2001; Almeida, 2002). Such techniques are worth to explore and integrate in the system for medical diagnosis and prediction.
Artificial neural networks (ANNs) are relatively new computational tools that have found extensive utilization in solving many complex real-world problems (Basheer and Hajmeer, 2000; Errejon, et. al., 2001; Almeida, 2002). The attractiveness of ANNs comes from their remarkable information processing characteristics pertinent mainly to nonlinearity, high parallelism, fault and noise tolerance, and learning and generalization capabilities (Basheer and Hajmeer, 2000). Artificial neural networks (ANNs) are a type of artificial intelligence software inspired by biological neuronal systems that can be used for nonlinear statistical modeling (Errejon, Crawford, Dayhoff, O’Donnell, Tewari, Finkelstein, and Gamito, 2001). In recent years, these applications have played an increasing role in predictive and classification modeling in medical research. ANNs have been applied to an increasing number of world problems of considerable complexity. Their most important advantage is in solving problems that are too complex for conventional technologies – problems that do not have an algorithmic solution or for which an algorithmic solution is too complex to be found. In general, because of their abstraction from the biological brain,
Artificial neural networks make a highly specialized tools in data transformation. The human brain has become an inspiration for the makers of artificial neural networks. Although even though artificial neural networks are more frequently used in areas like financial analysis, marketing studies or economical modeling, their application in psychology and medicine has given a lot of promising and fascinating discoveries.
Artificial neural networks (ANNs) are biologically inspired computer programs designed to simulate the way in which the human brain processes information (Agatonovic-Kustrin and Beresford, 2000). ANNs gather their knowledge by detecting the patterns and relationships in data and learn (or are trained) through experience, not from programming. An ANN is formed from hundreds of single units, artificial neurons or processing elements (PE), connected with coefficients (weights), which constitute the neural structure and are organized in layers (Agatonovic-Kustrin and Beresford, 2000). The power of neural computations comes from connecting neurons in a network.
A simple associationist neural network learns to factor abstract rules (i.e., grammars) from sequences of arbitrary input symbols by inventing abstract representations that accommodate unseen symbol sets as well as unseen but similar grammars (Hanson and Negishi, 2002). The neural network is shown to have the ability to transfer grammatical knowledge to both new symbol vocabularies and new grammars (Hanson and Negishi, 2002). Analysis of the state-space shows that the network learns generalized abstract structures of the input and is not simply memorizing the input strings. These representations are context sensitive, hierarchical, and based on the state variable of the finite-state machines that the neural network has learned (Hanson and Negishi, 2002).
On the other hand, associative neural network (ASNN) represents a combination of an ensemble of feed-forward neural networks and the k-nearest neighbor technique (Tetko, 2002). This method uses the correlation between ensemble responses as a measure of distance amid the analyzed cases for the nearest neighbor technique. This provides an improved prediction by the bias correction of the neural network ensemble (Tetko, 2002). An associative neural network has a memory that can coincide with the training set. If new data becomes available, the network further improves its predictive ability and provides a reasonable approximation of the unknown function without a need to retrain the neural network ensemble. This feature of the method dramatically improves its predictive ability over traditional neural networks and k-nearest neighbor techniques, as demonstrated using several artificial data sets and a program to predict lipophilicity of chemical compounds. (Tetko, 2002).
An artificial neural network (ANN) is an artificial intelligence tool that identifies arbitrary nonlinear multiparametric discriminant functions directly from experimental data (Almeida, 2002). The use of ANNs has gained increasing popularity for applications where a mechanistic description of the dependency between dependent and independent variables is either unknown or very complex. This machine learning technique can be roughly described as a universal algebraic function that will distinguish signal from noise directly from experimental data (Almeida, 2002). The application of ANNs to complex relationships makes them highly attractive for the study of biological systems (Almeida, 2002).
History of Artificial Neural Networks
The history of neural networks begins with the earliest model of the biological neuron given by McCulloch and Pitts in 1943. This model describes a neuron as a linear threshold computing unit with multiple inputs and a single output of either 0, if the nerve cell remains inactive, or 1, if the cell fires (McCulluch and Pitts, 1943; Stern, 2000).
The McCulloch and Pitts model was utilized in the development of the first artificial neural network by Rosenblatt in 1959 (Fraser, 2000). This network was based on a unit called the perceptron, which produces an output scaled as 1 or -1 depending upon the weighted, linear combination of inputs. Variations on the perceptron-based artificial neural network were further explored during the 1960s by Rosenblatt (1962) and by Widrow and Hoff (1960), among others.
In 1969 Minsky and Papert demonstrated that the perceptron was incapable of representing simple functions which were linearly inseparable. Because of this fundamental flaw (and Rosenblatt’s untimely death) the study of neural networks fell into something of a decline during the 1970s. However, once this limitation was overcome in the early 1980s (Fraser, 2000).
A distinction is to be made between the term “neural network” and “artificial neural network.” “Neural network” is often used to indicate networks that are hard-wired structures. “Artificial neural network” generally refers to a software-based or simulated structure; an algorithm rather than a physical system. While slow by comparison, ANNs are generally much cheaper to design and implement than their solid-state counterparts (Scaffer, 1997; Fraser, 2000).
Several different artificial neural network architectures have been developed over the past two decades (Fraser, 2000). Some of these represent elaborations of the simpler perceptron while others have arisen by integrating ideas from probability and fuzzy logic (Fraser, 2000). Common to all of these architectures is the linking together of many neurons, or nodes, into numerous interconnected pathways of input and output. By subjecting an ANN to various supervised or unsupervised training paradigms, the network “learns” to generalize a correct predictive (or associative) response to a given data set.
Since the early nineties the number of scientific papers dealing with the applications of artificial neural networks (ANNs) in medicinal chemistry and medicine fields has been dramatically increasing (Ochoa, Chana and Stud, 2001). Ochoa, Chana and Stud (2001) reviewed the applications of artificial neural networks in QSAR, concerning the classification or prediction of a biological activity, conformation searching, receptor docking and molecular design, are reviewed over the last five years. The different models of neural network used, the selection of descriptors, the comparison of results obtained by employing ANNs and other computational methods in QSAR studies, showed that there was an improvement of imaging information(Ochoa, Chana and Stud, 2001).
Uses of Artificial Neural Networks
ANN and Cancer Detection and Treatment
There is a great need for accurate treatment and outcome prediction in cancer. Two methods for prediction, artificial neural networks and Kaplan–Meier plots was compared by Bostwick and Burke (2001) for treatment and outcome prediction in cancer. Boswick and Burke (2001) proposed that artificial neural networks are useful for prediction of outcome for individual patients with cancer because they are as accurate as the best traditional statistical methods, are able to capture complex phenomena without a priori knowledge, and can be reduced to a simpler model if the phenomena are not complex. On the other hand, Kaplan–Meier plots are of limited accuracy for prediction because they require partitioning of variables, require cutting continuous variables into discrete pieces, and can only handle one or two variables effectively (Bostwick and Burke, 2001). Artificial neural networks are an efficient statistical method for outcome prediction in cancer that utilizes all available powerful prognostic factors and maximizes predictive accuracy. Use of Kaplan–Meier plots for predictions is discouraged because of serious technical limitations and low accuracy (Bostwick and Burke, 2001).
In addition, it is important to predict outcome for colorectal cancer patients following surgery, as almost 50% of patients undergoing a potentially curative resection will experience relapse (Grumett, Snow and Kerr, 2003). Grumett, Snow and Kerr (2003) argued that the present prognostic categories such as Dukes or TNM staging are too broad, and further refining is required to prognosticate for high-risk subgroups. The authors argued that the neural network approach is superior to regression analysis.
Intelligent computerized systems can provide useful assistance to the physician in the rapid identification of tissue abnormalities and accurate diagnosis in real-time. Karkanis, Magoulas and Theofanous (2000) reviewed basic issues in medical imaging and neural network-based systems for medical image interpretation. In the framework of intelligent systems, a simple scheme that has been implemented is presented as an example of the use of intelligent systems to discriminate between normal and cancerous regions in colonoscopic images (Karkanis, Magoulas and Theofanous, 2000). Preliminary results indicate that this scheme is capable of high accuracy detection of abnormalities within the image. It can also be successfully applied to different types of images, to detect abnormalities that belong to different cancer types (Karkanis, Magoulas and Theofanous, 2000).
Knowledge-based decision support systems traditionally rely on condition-action rule structures, an adequate representation for simple decisions. In complex domains an important part of decision-making includes analysis of the consequences of a decision. Consequential reasoning is particularly important in medicine as potential risk and/or benefit can be included in ANN (Hudson, Cohen, and Hudson. 2001). The result is a method that is sensitive to individual patient reactions to chemotherapy agents, permitting an individualized approach to therapy (Hudson, Cohen, and Hudson. 2001). Individualized drug therapy is becoming increasingly feasible due to advances made in the field of genomics. The system is structured so that new information can be incorporated easily. Although the application shown is to chemotherapy, the general methodology can be used in any area in which the consequences should significantly influence the decision (Hudson, Cohen, and Hudson. 2001).
Khan, et. al. (2001) developed a method of classifying cancers to specific diagnostic categories based on their gene expression signatures using artificial neural networks (ANNs) by training the ANNs using the small, round blue-cell tumors (SRBCTs) as a model. The cancers belong to four distinct diagnostic categories and often present diagnostic dilemmas in clinical practice. The result showed that the ANNs correctly classified all samples and identified the genes most relevant to the classification. Furthermore, to test the ability of the trained ANN models to recognize SRBCTs, the authors analyzed additional blinded samples that were not previously used for the training procedure, and correctly classified them in all cases. Their study demonstrated the potential applications of these methods for tumor diagnosis and the identification of candidate targets for therapy (Khan, et. al., 2001).
Uses of ANN in Breast Cancer
Accurate and reliable decision making in breast cancer prognosis can help in the planning of suitable surgery and therapy and, generally, optimize patient management through the different stages of the disease. In recent years, several prognostic factors have been used as indicators of disease progression in breast cancer (Seker et. al., 2002).
Moreover, the prediction of clinical outcome of patients after breast cancer surgery plays an important role in medical tasks such as diagnosis and treatment planning (Jerez-Aragones, Gomez-Ruiz, Ramos-Jimenez, Munoz-Perez and Alba-Conejo, 2003). Different prognostic factors for breast cancer outcome appear to be significant predictors for overall survival, but probably form part of a bigger picture comprising many factors (Jerez-Aragones et.al, 2003). Survival estimations are currently performed by clinicians using the statistical techniques of survival analysis. In this sense, artificial neural networks are shown to be a powerful tool for analyzing datasets where there are complicated non-linear interactions between the input data and the information to be predicted (Jerez-Aragones et.al, 2003).
On the other an experiment by Seker, et. el (2002) showed otherwise. Nodal involvement and survival analyses in breast cancer are carried. Seker, et al (2002) predicted that this technique has produced more reliable prognostic factor models than those obtained using either the statistical or artificial neural networks-based methods.
Jerez-Aragones et.al (2003) presented a decision support tool for the prognosis of breast cancer relapse that combines a novel algorithm TDIDT (control of induction by sample division method, CIDIM), to select the most relevant prognostic factors for the accurate prognosis of breast cancer, with a system composed of different neural networks topologies that takes as input the selected variables in order for it to reach good correct classification probability. Jerez-Aragones et.al (2003) asserted that the proposed system is an useful tool to be used by clinicians to search through large datasets seeking subtle patterns in prognostic factors, and that may further assist the selection of appropriate adjuvant treatments for the individual patient.
In a separate study. Hung, Shanker and Hu (2002) reported the results of using neural networks for breast cancer diagnosis. The theoretical advantage is that posterior probabilities of malignancy can be estimated directly, and coupled with resampling techniques such as the bootstrap, distributions of the probabilities can also be obtained. The result supports the studies and asserts that artificial neural networks best predicts breast cancer.
Abbass (2002) presented an evolutionary artificial neural network (EANN) approach based on the pareto-differential evolution (PDE) algorithm augmented with local search for the prediction of breast cancer. The approach artificial neural networks (ANNs) according to Abbass (2002) could be used to improve the work of medical practitioners in the diagnosis of breast cancer. Their abilities to approximate nonlinear functions and capture complex relationships in the data are instrumental abilities which could support the medical domain.
Uses of ANN in Skin Cancer
Successful treatment of skin cancer, especially melanoma, depends on early detection, but diagnostic accuracy, even by experts, can be as low as 56% so there is an urgent need for a simple, accurate, non-invasive diagnostic tool (Wallace, Bamber, Crawford, Ott, and Mortimer, 2000). Wallace, Bamber, Crawford, Ott, and Mortimer (2000) compared the performance of an artificial neural network (ANN) and multivariate discriminant analysis (MDA) for the classification of optical reflectance spectra from malignant melanoma and benign naevi. The ANN was significantly better than MDA, especially when a larger data set was used, where the classification accuracy was 86.7% for ANN and 72.0% for MDA. ANN was better at learning new cases than MDA for this particular classification task. This study has confirmed that the convenience of ANNs could lead to the medical community and patients benefiting from the improved diagnostic performance which can be achieved by objective measurement of pigmented skin lesions using spectrophotometry (Wallace, Bamber, Crawford, Ott, and Mortimer, 2000).
Uses of ANN in Prostate Cancer
Artificial neural networks (ANNs) have only recently been applied to solve problems in the diagnosis, staging, and prediction of treatment outcome in prostate cancer (Babaian and Zhang, 2001). Babaian and Zhang (2001) suggested that the continued development and refinement of computer-assisted diagnostic methodology are warranted to enhance conventional statistical approaches to the classification and pattern recognition found in data sets from men either suspected of having or known to have prostate cancer.
Furthermore, Errejon, et. al. (2001) reviewed the basic concepts behind ANNs and examined the role of this technology in selected applications in prostate cancer research. Their study showed that there is a higher probability of accurate diagnosis when the ANN software is used.
Uses of ANN in Neurobiology
The history of the evolution of neurocomputing and its relation to the field of neurobiology, with special emphasis on back propagation (BP) ANNs theory and design shows that the most common problems that BPANNs developers face during training can be used to solve microbial growth curves of S. flexneri. The developed model was reasonably accurate in simulating both training and test time-dependent growth curves as affected by temperature and pH (Basheer and Hajmeer, 2000).
In recent years, data from neurobiological experiments have made it increasingly clear that biological neural networks, which communicate through pulses, use the timing of the pulses to transmit information and perform computation (Maas and Bishop, 2001). This realization has stimulated significant research on pulsed neural networks, including theoretical analyses and model development, neurobiological modeling, and hardware implementation (Maas and Bishop, 2001). Gluck and Myers (2001) emphasized the function of brain structures as they give rise to behavior, rather than the molecular or neuronal details.
The behavior of a neural network is determined by the transfer functions of its neurons, by the learning rule, and by the architecture itself (Agatonovic-Kustrin and Beresford, 2000). Agatonovic-Kustrin and Beresford (2000) explained that the weights are the adjustable parameters and, in that sense, a neural network is a parameterized system. The weighed sum of the inputs constitutes the activation of the neuron. The activation signal is passed through transfer function to produce a single output of the neuron. Transfer function introduces non-linearity to the network. During training, the inter-unit connections are optimized until the error in predictions is minimized and the network reaches the specified level of accuracy. Thus, the author surmised that once the network is trained and tested it can be given new input information to predict the output.
For instance, dementia often reduces a person’s ability to perform activities of daily living because he or she becomes confused and cannot remember the sequence of steps to perform. The current solution is to have a caregiver continually supervise and assist the person using verbal reminders or cues. This loss of privacy and increased dependency may cause the affected person to become embarrassed and agitated. We propose that this situation might be improved by using a computerized device that monitors progress and provides the reminders needed. The COACH is a first prototype of such a device. It uses artificial intelligence to observe a user, learn from his or her actions, and issue prerecorded cues of varying detail (Mihailidis, Fernie and Barbenel, 2001). The device was developed using a personal computer and a video camera that unobtrusively tracked the user. Preliminary testing with subjects who simulated confused behavior as they washed their hands showed that the device was performing its functions with an efficacy of approximately 95% (Mihailidis, Fernie and Barbenel, 2001).
Localization of focal electrical activity in the brain using dipole source analysis of the electroencephalogram (EEG), is usually performed by iteratively determining the location and orientation of the dipole source, until optimal correspondence is reached between the dipole source and the measured potential distribution on the head. Van Hoey et. al., (2000) investigated the use of feed-forward layered artificial neural networks (ANNs) to replace the iterative localization procedure, in order to decrease the calculation time. The localization accuracy of the ANN approach is studied within spherical and realistic head models.
Uses of ANN in Genetic Analysis
Correct identification of the Translation Initiation Start (TIS) in cDNA sequences is an important issue for genome annotation. Hatzigeorgiou (2002) improved upon current methods and provide a performance guaranteed prediction. This is achieved by using two modules, one sensitive to the conserved motif and the other sensitive to the coding/non-coding potential around the start codon. Both modules are based on Artificial Neural Networks (ANNs). By applying the simplified method of the ribosome scanning model, the algorithm starts a linear search at the beginning of the coding ORF and stops once the combination of the two modules predicts a positive score. The result showed that TIS were correctly predicted.
Thalassemias are pathologies that derive from genetic defects of the globin genes. The most common defects among the population affect the genes that are involved in the synthesis of alpha and beta chains. The main aspects of these pathologies are well explained from a biochemical and genetic point of view (Amendolia, et al., 2002). The diagnosis is fundamentally based on hematologic and genetic tests. A genetic analysis is particularly important to determine the carriers of alpha-thalassemia, whose identification by means of the hematologic parameters is more difficult in comparison with heterozygotes for alpha-thalassemia (Amendolia, et al., 2002).
Amendiola, et. al. (2002) investigated the use of artificial neural networks (ANNs) for the classification of thalassemic pathologies using the hematologic parameters resulting from hemochromocytometric analysis only. On the basis of their results, an automated system that allows real-time support for diagnoses was proposed.
ANN and Oncological Prognosis
Accurate and reliable decision making in oncological prognosis can help in the planning of suitable surgery and therapy, and generally, improve patient management through the different stages of the disease. In recent years, several prognostic markers have been used as indicators of disease progression in oncology. However, the rapid increase in the discovery of novel prognostic markers resulting from the development in medical technology, has dictated the need for developing reliable methods for extracting clinically significant markers where complex and nonlinear interactions between these markers naturally exist (Seker, Odetayo, Petrovic and Naguib 2003).
Seker, Odetayo, Petrovic and Naguib (2003) explored the fuzzy k-nearest neighbor (FK-NN) classifier as a fuzzy logic method that provides a certainty degree for prognostic decision and assessment of the markers, and to compare it with: 1) logistic regression as a statistical method and 2) multilayer feedforward back propagation neural networks an artificial neural-network tool, the latter two techniques having been widely used for oncological prognosis. In order to achieve this aim, breast and prostate cancer data sets are considered as benchmarks for this analysis. The overall results obtained indicate that the Neural Network-based method yields the highest predictive accuracy, and that it has produced a more reliable prognostic marker model than the statistical and artificial neural-network-based methods (Seker, Odetayo, Petrovic and Naguib 2003).
The application of artificial neural networks (ANNs) for prognostic and diagnostic classification in clinical medicine has become very popular (Schwarzer, Vach and Schumacher, 2000). In particular, feed-forward neural networks have been used extensively, often accompanied by exaggerated statements of their potential. However, Schwarzer, Vach and Schumacher (2000) point out that the uncritical use of ANNs may lead to serious problems, such as the fitting of implausible functions to describe the probability of class membership and the underestimation of misclassification probabilities. In applications of ANNs to survival data, further difficulties arise. Finally, the results of a search in the medical literature from 1991 to 1995 on applications of ANNs in oncology and some important common mistakes are reported (Schwarzer, Vach and Schumacher, 2000).
Uses of ANN in Cardiovascular Treatment
Folland, Hines, Boilot and Morgan (2002) used an artificial neural networks (ANNs) to the analysis of heart sound abnormalities through auscultation. They predicted that the application of ANNs to the analysis of respiratory auscultation and consequently the development of a combined cardio-respiratory analysis system using auscultated data could lead to faster and more efficient treatment.
An automated online technique is described by Newey and Nassiri (2002) for measurement of artery diameter in flow-mediated dilation (FMD) ultrasound (US) images, using artificial neural networks to identify and track artery walls. This allows FMD results to be calculated without the inherent delay of current retrospective methods (Newey and Nassiri, 2002). Advantages of the technique include: online analysis; wall tracking optimization before the study proper; measurement of diameter changes over the cardiac cycle; low FMD measurement variance; minimal image degradation; and no unwieldy image store (Newey and Nassiri, 2002). Measurement of artery diameter changes over the cardiac cycle was explored using simulated image sequences generated with a virtual US scanner (Newey and Nassiri, 2002).
ANN and Chronic Fatigue and Pulmonary Embolism
The definition of chronic fatigue syndrome (CFS) is still disputed and no validated classification criteria have been published. Artificial neural networks (ANN) are computer-based models that can help to evaluate complex correlations was proposed by Linder, Dinser, Wagner, Kruger and Hoffman (2002) to reduce and cure CFS. Classification criteria developed by ANN were found to have a sensitivity of 95% and a specificity of 85%. ANN achieved a higher accuracy than any of the other methods. Thus, the most accurate criteria were derived with the help of ANN. We therefore recommend the use of ANN for the classification of syndromes with complex interrelated symptoms like CFS (Linder, Dinser, Wagner, Krueger and Hoffmann, 2002).
Eng (2002) determined whether an artificial neural network, a new data analysis method, offers increased performance over conventional logistic regression in predicting the presence of a pulmonary embolism for patients in a well-known data set. In the study population, the usefulness of data from ventilation-perfusion scans as predictors of the presence of a pulmonary embolism was similar for the three analytic methods, a finding that reinforces the importance of making comparisons to simpler or more established methods when performing studies involving complex analytic models, such as artificial neural networks (Eng, 2002).
Uses of ANN in ICUs and Sound Detection in ICUs
Expert systems are used widely to cross reference symptoms and diseases and therefore greatly improve the accuracy of diagnostics. Object recognition can be used in images from cats cans or x-ray machines to get preliminary analysis of these images (AI, 2000). AI enabled technology is also used in monitoring and control in intensive care units in hospitals.
Small changes that occur in a patient’s physiology over long periods of time are difficult to detect, yet they can lead to catastrophic outcomes. Detecting such changes is even more difficult in intensive care unit (ICU) environments where clinicians are bombarded by a barrage of complex monitoring signals from various devices. Early detection accompanied by appropriate intervention can lead to improvement in patient care. Neural networks can be used as the basis for an intelligent early warning system (Parmanto, Deneault, and Denault, 2001). Parmanto, Deneault, and Denault (2001) developed a time-delay neural networks (TDNN) for classifying and detecting hemodynamic changes. The TDNN were trained with these matrices and subsequently tested to predict unseen cases. The TDNN perform remarkably not only in identifying all hemodynamic conditions, but also in quickly detecting their changes (Parmanto, Deneault, and Denault, 2001).
Frize, Ennett, Stevenson, and Trigg (2001) provided an overview of applications of artificial neural networks (ANNs) to various medical problems, with a particular focus on the intensive care unit environment (ICU). Several technical approaches were tested to see whether they improve the ANN performance in estimating medical outcomes and resource utilization in adult ICUs. These experiments include: (1) use of the weight-elimination cost function; (2) use of ‘high’ and ‘low’ nodes for input variables; (3) verifying the effect of the total number of input variables on the results; (4) testing the impact of the value of the constant predictor on the performance of the ANNs (Frize, Ennett, Stevenson, and Trigg, 2001). The developments Frize, Ennett, Stevenson, and Trigg (2001) argued can help medical and nursing personnel to assess patient status, assist in making a diagnosis, and facilitate the selection of a course of therapy.
In earlier work, the research group by Tong, Frize and Walker (2002) successfully used artificial neural networks (ANNs) to estimate ventilation duration for adult intensive care unit (ICU) patients. The ANNs performed well in terms of correct classification rate (CCR) and average squared error (ASE) classifying the outcome. Their new work applied this adult model to the estimation of ventilation with neonatal ICU (NICU) patient records. The performance obtained with the neonatal patients was comparable to that previously found with the adult database, again as measured in terms of a maximum CCR and a minimum ASE (Tong, Frize, and Walker, 2002). The effectiveness of using the weight-elimination technique in controlling over fitting was again validated for the neonatal patients as it had been for adult patients. It was concluded that the approach developed for ICU adult patients was also successfully applied to a different medical environment: neonatal ICU patients (Tong, Frize, and Walker, 2002).
In addition, (Van Hoey et al. (2000) investigate the robustness of both the iterative and the ANN approach by observing the influence on the localization error of both noise in the scalp potentials and scalp electrode mislocalizations. The study showed that the ANN localization approach appears to be robust to noise and electrode mislocations even in ICUs. In comparison with the iterative localization, the ANN provides a major speed-up of dipole source localization. (Van Hoey et al. 2000) concluded that an artificial neural network is a very suitable alternative for iterative dipole source localization in applications where large numbers of dipole localizations have to be performed, provided that an increase of the localization errors by a few millimeters is acceptable.
Moreover, the prediction of the length of stay at the moment of hospital admission of ICU patients is of utmost importance. Many studies have used lineal models to predict this variable, but there are inherent limitations to these models (Chacon, Rocco, Morgado, Saez and Pliscoff, 2002). Using ANN, the model determined that gastrointestinal diseases, infections and respiratory problems were the main causes of prolongation of intensive care unit stay (Chacon, Rocco, Morgado, Saez and Pliscoff, 2002).
Uses of ANN in Pharmaceuticals
Many types of neural networks have been designed already and new ones are invented every week but all can be described by the transfer functions of their neurons, by the learning rule, and by the connection formula (Agatonovic-Kustrin and Beresford, 2000). ANN represents a promising modeling technique, especially for data sets having non-linear relationships which are frequently encountered in pharmaceutical processes (Agatonovic-Kustrin and Beresford, 2000). In terms of model specification, artificial neural networks require no knowledge of the data source but, since they often contain many weights that must be estimated, they require large training sets.
In addition, ANNs can combine and incorporate both literature-based and experimental data to solve problems. The various applications of ANNs can be summarized into classification or pattern recognition, prediction and modeling (Agatonovic-Kustrin and Beresford, 2000).Supervised ‘associating networks can be applied in pharmaceutical fields as an alternative to conventional response surface methodology. The potential applications of ANN methodology in the pharmaceutical sciences range from interpretation of analytical data, drug and dosage form design through biopharmacy to clinical pharmacy (Agatonovic-Kustrin and Beresford, 2000).
For instance, the probabilistic neural networks (PNNs) is used for classification of bacterial growth/no-growth data and modeling the probability of growth. The PNN approach combines both Bayes theorem of conditional probability and Parzen’s method for estimating the probability density functions of the random variables (Hajmeer and Basheer, 2002). Unlike other neural network training paradigms, PNNs are characterized by high training speed and their ability to produce confidence levels for their classification decision (Hajmeer and Basheer, 2002). The PNN-based models were found by Hajmeer and Basheer (2002) to outperform linear and nonlinear logistic regression and MFANN in both the classification accuracy and ease by which PNN-based models are developed.
Moreover, antibiotic-resistant pathogens are increasingly prevalent in the hospitals and community (Shang, Lin, and Goetz, 2000). A timely and accurate diagnosis of the infection would greatly help physicians effectively treat patients. Shang, Lin, and Goetz (2000) investigated the potential of using neural networks (NN) and logistic regression (LR) approach in diagnosing methicillin-resistant Staphylococcus aureus (MRSA). The authors found out that Neural Networks is better than the logistic regression approach, in terms of both the discriminatory power and the robustness. With modeling flexibility inherent in its techniques, NN is effective in dealing with MRSA and other classification problems involving large numbers of variables and interaction complexity (Shang, Lin, and Goetz, 2000). On the other hand, logistic regression in our case is slightly inferior, offers more clarity and less perplexity. It could be a method of choice when fewer variables are involved and/or justification of the results is desired (Shang, Lin, and Goetz, 2000).
Fluoride Detection and ANN
Zhou, Yan, Xu, Wang, Chen and Hu (2000) dealt with the application of artificial neural networks (ANNs) to two common problems in spectroscopy: optimization of experimental conditions and non-linear calibration of the result, with particular reference to the determination of fluoride by flow injection analysis (FIA). The FIA system was based on the formation of a blue ternary complex between zirconium(IV), p-methyldibromoarsenazo and F- with the maximum absorption wavelength at 635 nm (Zhou, Yan, Xu, Wang, Chen and Hu, 2000). The result showed that good prediction was attained and the trained networks proved to be very powerful in all applications. The proposed method was successfully applied to the determination of free fluoride in tea and toothpaste (Zhou, Yan, Xu, Wang, Chen and Hu, 2000).
Issues in ANN
Clinical laboratories are being subjected to an ever-increasing workload (Dybowski, 2001). The complexity of patient-related data can be such that even an expert can overlook important details. With the production of high volumes of possibly high-dimensional data from instruments such as flow-cytographs, and the integration of disparate data sources through data fusion, this is becoming an increasing problem. Presented with such problems, it is natural that the medical fraternity should turn to ANNs in the hope that these adaptable models could alleviate at least some of the problems; however, the use of ANNs in medicine has raised a number of issues. One of these is the relationship between ANNs and statistics (Dybowski, 2001); another is the use of black box systems in medicine (AI, 2000).
A key issue in the use of neural network methods in a medical application of this kind is that it is unclear how the system reaches its decision. This ‘black box’ approach is not suited to the field of medicine, where it is important to make obvious how a decision was made, or a diagnosis reached (AI, 2000).
A criticism leveled against neural networks is that they are black-box systems (Dybowski, 2001; AI, 2000). By this it is meant that the manner in which a neural network derives an output value from a given feature vector is not comprehensible to the non-specialist, and that this lack of comprehension makes the output from neural networks unacceptable. According to Dybowski (2001) there are a number of properties the model, two of which are accuracy (the closeness of a model s estimated value to the true value) and interpretability. By interpretability, we mean the type of input-output relationships that can be extracted from a model and which are comprehensible to the intended users of the model.
Another approach to providing interpretability is to use a hybrid neuro- fuzzy system (Seker, Odetayo, Petrovic and Naguib, 2003). These are feed-forward networks built from if-then rules containing linguistic terms based on domain knowledge, and the membership functions associated with the fuzzy rules can be tuned to data by means of a training algorithm. Both rule extraction and hybrid neuro-fuzzy systems are responses to those clinicians unwilling to use a predictive model lacking interpretability, even when the model is highly accurate (Seker, Odetayo, Petrovic and Naguib, 2003).
Measuring Medical Predictability: ANN and Linear Models
The most common neural system for unsupervised training is Kohonen’s self-organizing feature maps (SOFMs) (Dybowski, 2001). The aim of SOFMs is to map an input vector to one of a set of neurons arranged in a lattice, and to do so in such a way that positions in input space are topologically ordered with locations on the lattice. The second most commonly used ANNs in medicine are the radial-basis function networks (RBFNs) (Dybowski, 2001). An RBFN can be regarded as a type of generalized linear discriminant function, a linear function of functions that permits the construction of non-linear decision surfaces. The basis functions of an RBFN define local responses (receptive fields). Typically, only some of the hidden units (basis functions)produce significant values for the final layers.
Medical prognosis has played an increasing role in health care. Reliable prognostic models that are based on survival analysis techniques have been recently applied to a variety of domains, with varying degrees of success (Machado, 2002). Machado (2002) reviewed some methods commonly used to model time-oriented data, such as Kaplan-Meier curves, Cox proportional hazards, and logistic regression, and discuss their applications in medical prognosis. Compared to nonlinear, nonparametric models such as neural networks have increasingly been used for building prognostic models.
Artificial neural networks (ANNs) are very popular as classification or regression mechanisms in medical decision support systems despite the fact that they are unstable predictors (Cunningham, Carney and Jacob, 2000). According to Cunningham, Carney and Jacob, (2000), this instability means that small changes in the training data used to build the model (i.e. train the ANN) may result in very different models. Moreover, a central implication of this is that different sets of training data may produce models with very different generalization accuracies. However, Cunningham, Carney and Jacob (2000) argued that the accuracy of such a predictor can be improved by aggregating the output of several predictors.
On the other hand, Ciampi and Zhang (2002) presented a new approach to training back-propagation artificial neural nets (BP-ANN) based on regularization and cross-validation and on initialization by a logistic regression (LR) model. The new approach is expected to produce a BP-ANN predictor at least as good as the LR-based one. The result, on the whole, confirm that, in practical situations of clinical interest, proper training may significantly improve the predictive performance of a BP-ANN (Ciampi and Zhang, 2002).
Significant advances have been made in recent years to improve measurement technology and performance of phosphor materials in the fields of optically stimulated luminescence (OSL) dosimetry (Lee, Kim, and Lee, 2001). In the process of the study conducted by Lee, Kim and Lee (2001), a new dose assessment algorithm was developed using artificial neural networks in hopes of achieving a higher degree of accuracy and precision in personal OSL dosimetry system. A feed forward neural network using the error back-propagation method with Bayesian optimization was applied for the response unfolding procedure. The validation of the proposed algorithm was investigated and was confirmed (Lee, Kim, and Lee, 2001).
Consequently, Price, et. al., (2000) introduced artificial neural networks (ANNs), flexible nonlinear modeling techniques that test a model’s generality by applying its estimates against “future” data. ANNs have potential for overcoming some shortcomings of linear models. The basics of ANNs and their applications to psychological assessment are reviewed, the study showed that ANN had a relatively higher degree of accuracy in predicting results than the linear models.
ANN and Black Boxes
Researchers who design intelligent systems for medical decision support, are aware of the need for response to real clinical issues, in particular the need to address the specific ethical problems that the medical domain has in using black boxes (Smith, Nugent, and McClean, 2003). This means such intelligent systems have to be thoroughly evaluated, for acceptability. Attempts at compliance, however, are hampered by lack of guidelines (Smith, Nugent, and McClean, 2003).
Using artificial neural networks (ANNs) in medical applications can be challenging because of the often-experimental nature of ANN construction and the “black box” label that is frequently attached to them (Rodvold, 2001). In the US, medical neural networks are regulated by the Food and Drug Administration.
References
Abbass, HA. (July 2002) An evolutionary artificial neural networks approach for breast cancer diagnosis. Artif Intell Med. 25(3):265-81.
Agatonovic-Kustrin S. and Beresford, R. (June 2000) Basic concepts of artificial neural network (ANN) modeling and its application in pharmaceutical research. J Pharm Biomed Anal. 22(5):717-27.
Almeida, JS. (Feb 2002) Predictive non-linear modeling of complex data by artificial neural networks. Curr Opin Biotechnol. 13(1):72-6.
Amendolia SR, Brunetti A, Carta P, Cossu G, Ganadu ML, Golosio B, Mura GM, and Pirastru MG. (Jan-Feb 2002). A real-time classification system of thalassemic pathologies based on artificial neural networks. Med Decis Making. 22(1):18-26
Artificial Intelligence. (2000) An Illustrative Overview, AISB Quarterly, Winter, No. 103, Pages 31-35
Babaian RJ, Zhang Z. (2001). Computer-assisted diagnostics: application to prostate cancer. Mol Urol. Winter;5(4):175-80.
Basheer IA, Hajmeer M. (Dec 2000). Artificial neural networks: fundamentals, computing, design, and application. J Microbiol Methods. 43(1):3-31.
Bostwick DG and Burke HB. (April 2001). Prediction of individual patient outcome in cancer: comparison of artificial neural networks and Kaplan–Meier methods. Cancer. 15;91(8 Suppl):1643-6.
Brooks C. and Tsolacos S. (June 2003). International evidence on the predictability of returns to securitized real estate assets: econometric models versus neural networks. Journal of Property Research. Vol. 20, No. 2. pp. 133 155(23)
Ciampi A. and Zhang F. (May 2002). A new approach to training back-propagation artificial neural networks: empirical evaluation on ten data sets from clinical studies. Stat Med. 15;21(9):1309-30.
Chacon M, Rocco V, Morgado E, Saez E, and Pliscoff S. (Jan 2002) Prediction of the length of stay in Intensive Care Units using artificial neural networks. Rev Med Chil. 130(1):71-8.
Cunningham P, Carney J, Jacob S. (Nov 2000) Stability problems with artificial neural networks and the ensemble solution. Artif Intell Med. 2000;20(3):217-25.
Dybowski, R. (2001) Neural Computation in Medicine: Perspectives and Prospects. In Malmgren H, Borga M, Niklasson L. (eds.) Proceedings of the ANNIMAB-1 Conference (Artificial Neural Networks in Medicine and Biology), Goteborg, 13-16 May 2000 (Springer, 2000) pp. 26 36.
Eng, J. (Oct 2002) Predicting the presence of acute pulmonary embolism: a comparative analysis of the artificial neural network, logistic regression, and threshold models. AJR Am J Roentgenol. 179(4):869-74.
Errejon A, Crawford ED, Dayhoff J, O’Donnell C, Tewari A, Finkelstein J, and Gamito EJ. (2001) Use of artificial neural networks in prostate cancer. Mol Urol. Winter;5(4):153-8.
Folland R, Hines EL, Boilot P, and Morgan D. (May 2002) Classifying coronary dysfunction using neural networks through cardiovascular auscultation. Med Biol Eng Comput. 40(3):339-43.
Fraser, C. (2000) Neural Networks: Literature Review From a Statistical Perspective. California State University, Hayward — Spring.
Frize M, Ennett CM, Stevenson M, and Trigg HC. (April 2001) Clinical decision support systems for intensive care units: using artificial neural networks. Med Eng Phys. 23(3):217-25
Gluck, M. and Myers, C. (August 2001) Gateway to Memory. MIT Press. 464 pp.
Grumett S, Snow P, and Kerr D. (Feb 2003) Neural networks in the prediction of survival in patients with colorectal cancer. Clin Colorectal Cancer. 2(4):239-44
Hajmeer M, Basheer I. (Oct 2002) A probabilistic neural network approach for modeling and classification of bacterial growth/no-growth data. J Microbiol Methods. 51(2):217-26.
Hanson SJ, and Negishi M. (Sept 2002). On the emergence of rules in neural networks. Neural Comput. 14(9):2245-68.
Hatzigeorgiou A.G. (Feb 2002) Translation initiation start prediction in human cDNAs with high accuracy. Bioinformatics. Vol. 18, No. 2. pp. 343-350(8).
Hudson DL, Cohen ME, and Hudson SE. (2001) The use of consequential reasoning in cancer chemotherapy. Medinfo. 10(Pt 2):1349-53.
Hung M.S., Shanker M., and Hu, M.Y. (Feb 2002) Estimating breast cancer risks using neural networks. Journal of the Operational Research Society. Vol. 53, No. 2. pp. 222-231(10)
Ishak, W.H. and Siraj, F. (2002). Artificial Intelligence in Medical Application: An Exploration. Health Informatics
Jerez-Aragones JM, Gomez-Ruiz JA, Ramos-Jimenez G, Munoz-Perez J, and Alba-Conejo E. (Jan 2003). A combined neural network and decision trees model for prognosis of breast cancer relapse. Artif Intell Med. 27(1):45-63.
Karkanis S., Magoulas G.D. and Theofanous N. (2000) Image recognition and neuronal networks: intelligent systems for the improvement of imaging information. Minimally Invasive Therapy and Allied Technologies. Vol. 9, No. 3-4, pp. 225-230(6).
Khan J, Wei JS, Ringner M, Saal LH, Ladanyi M, Westermann F, Berthold F, Schwab M, Antonescu CR, Peterson C, and Meltzer PS. (Jun 2001). Classification and diagnostic prediction of cancers using gene expression profiling and artificial neural networks. Nat Med. 7(6):673-9.
Lee SY, Kim BH, and Lee KJ. (June 2001). An application of artificial neural intelligence for personal dose assessment using a multi-area OSL dosimetry system. Radiat Meas. 33(3):293-304.
Linder R, Dinser R, Wagner M, Krueger GR, and Hoffmann A. (Jan- Feb 2002). Generation of classification criteria for chronic fatigue syndrome using an artificial neural network and traditional criteria set. In Vivo. 16(1):37-43.
Maass, W. and Bishop, C. (March 2001) Pulsed Neural Networks. MIT Press. Pp. 407.
Machado, O. (May 2002) Artificial neural networks for diagnosis and prognosis in prostate cancer. Semin Urol Oncol.20(2):89-95.
McCulloch, W.S., and Pitts, W. (1943). A Logical Calculus of the Ideas Immanent in Nervous Activity. Bulletin of Mathematical Biophysics, 5, 115-133.
Mihailidis A, Fernie GR, and Barbenel JC. (2001). The use of artificial intelligence in the design of an intelligent cognitive orthosis for people with dementia. Assist Technol. 13(1):23-39.
Newey VR and Nassiri DK. (Feb 2002). Online artery diameter measurement in ultrasound images using artificial neural networks. Ultrasound Med Biol. 28(2):209-16.
Ochoa, C., Chana, A. and Stud, M. (Nov 2001) Applications of Neural Networks in the Medicinal Chemistry Field. Current Medicinal Chemistry-Nervous System Agents. Vol. 1, No. 3, pp. 247-256(10).
Parmanto B, Deneault LG, and Denault AY. (Sept 2001) Detection of hemodynamic changes in clinical monitoring by time-delay neural networks. Int J Med Inf. 63(1-2):91-9.
Price RK, Spitznagel EL, Downey TJ, Meyer DJ, Risk NK, and el-Ghazzawy OG. (March 2000). Applying artificial neural network models to clinical decision making. Psychol Assess. 12(1):40-51.
Rodvold, DM. (2001) Validation and regulation of medical neural networks. Mol Urol. Winter;5(4):141-5.
Rosenblatt, F. (1959). The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain. Psychological Review 65:386-408.
Rosenblatt, F. (1962). Principles of Neurodynamics, New York: Spartan Books.
Schalkoff, R.J. (1997), Artificial Neural Networks, New York: McGraw-Hill.
Seker H, Odetayo MO, Petrovic D, and Naguib RN. (Jun 2003). A fuzzy logic based-method for prognostic decision making in breast and prostate cancers. IEEE Trans Inf Technol Biomed. 7(2):114-22.
Seker H, Odetayo MO, Petrovic D, Naguib RN, Bartoli C, Alasio L, Lakshmi MS, and Sherbet GV. (Jan-Feb 2002). Assessment of nodal involvement and survival analysis in breast cancer patients using image cytometric data: statistical, neural network and fuzzy approaches. Anticancer Res. 22(1A):433-8.
Shang JS, Lin YS, and Goetz AM. (Sept 2000) Diagnosis of MRSA with neural networks and logistic regression approach. Health Care Manag Sci. 3(4):287-97
Smith AE, Nugent CD, and McClean SI. (Jan 2003). Evaluation of inherent performance of intelligent medical decision support systems: utilising neural networks as an example. Artif Intell Med. 27(1):1-27.
Stern, H.S. (2000). Neural Networks in Applied Statistics. Technometrics, 38, 205-214.
Tetko IV. (May 2002) Neural network studies. 4. Introduction to associative neural networks. J Chem Inf Comput Sci. 42(3):717-28.
Tong Y, Frize M, and Walker R. (June 2002). Extending ventilation duration estimations approach from adult to neonatal intensive care patients using artificial neural networks. IEEE Trans Inf Technol Biomed. 6(2):188-91.
Van Hoey G, De Clercq J, Vanrumste B, Van De Walle R, Lemahieu I, D’Have M, and Boon P. (April 2000). EEG dipole source localization using artificial neural networks. Phys Med Biol. 45(4):997-1011.
Wallace VP, Bamber JC, Crawford DC, Ott RJ, and Mortimer PS. (Oct 2000) Classification of reflectance spectra from pigmented skin lesions, a comparison of multivariate discriminant analysis and artificial neural networks. Phys Med Biol. 45(10):2859-71.
Widrow, B., and Hoff, M.E. (1960). Adaptive Switching Circuits. IRE WESCON Convention Record, 4, 96-104.
Zhou Y, Yan A, Xu H, Wang K, Chen X, and Hu Z. (Dec 2000) Flow injection analysis of fluoride: optimization of experimental conditions and non-linear calibration using artificial neural networks. Analyst. 125(12):2376-80.
Credit:ivythesis.typepad.com
0 comments:
Post a Comment