Applications of Connectionist Systems in Biomedicine Sabbatini R.M.E. Center for Biomedical Informatics, State University of Campinas, P.O. Box 6005, 13081 Campinas, S<176>o Paulo, Brazil. ABSTRACT Artificial neural networks, or connectionist systems, are being increasingly used to represent and to processs information by means of networks of interconnected processing elements, similar to neurons. Several emerging global properties of connectionist systems, such as associative memory, distributed parallel processing, learning, etc.; have favored its applications in a large variety of tasks involving pattern classification and recognition. Connectionist systems constitute a new and interesting paradigm for the area of Artificial Intelligence in Medicine, and have found applications in processing and interpretation of biological signals and images, decision support systems, etc., substituting advantageously many conventional techniques based on statistical classification and logical systems. This paper discusses the potential applications and benefits of connectionist systems in Biology and Medicine and reviews the field in its latest developments. 1. Introduction The development of Artificial Intelligence in the last years has been characterized by a remarkable growth in interest and number of investigations around a theoretical model which, having been almost abandoned in the sixties, has witnessed a vigorously rebirth in the eighties, on the basis of new and important discoveries ane inventions. The connectionist model, also interchangeably called as neuromorphic, or simply, artificial neural network (ANN) systems, arose in the forties with the first studies on the formal organization of the neural tissue. Pioneering theoretical studies, such as those by McCulloch and Pitts [1] led to a new model of computational structure based on the concept of cooperative processing, carried out by a network of binary elements, which is able to execute all kinds of elementary logical operations required for building a computer. One of the advantages of this new kind of computer in relation to conventional sequential processors is the possibility of achieving true parallel and distributed processing, similar to the brain, with an evident increase in speed. Another important feature of neural-like networks, studied later on, is that they exhibit self-organizing behavior, that is, they "learn" new associative and classificatory tasks by means of embedded adaptive mechanisms. From these studies were born the first practical models, such as the perceptrons [2], which showed surprising performance in complex tasks of pattern recognition and classification. An important, related concept was that of associative memory, whereby information is spread about a network in a 'holographic' manner, its retrieval being possible by means of contents addressing [3]. However, due to several intervening factors, such as the growth of Artificial Intelligence in the directon of logical and heuristic systems, triggered in the seventies by a wide dissemination and use of von Neumann computers, by the appearance of powerful symbolic languages, such as LISP, and by the apparent success of rule-based expert systems; as well as by some theoretical objections to artificial neural networks [4]; caused a strong inhibition in research and development on connectionist systems, which lasted more than two decades. The interest for this model, however, started again to grow exponentially, due mostly to important theoretical and practical advances in the eighties [5]. An avalanche of new architectures, algorithms and applications followed. On the other hand, the possibility of building computers with processing speeds several magnitudes higher than sequential processors, using parallel machines, and the recognition that certain tasks are better processed by brain-like structures (such as speech recognition, vision, etc.), has led, in the nineties, to the rapid development of the first prototype computers and processors/memories with neuromorphic structure, the so-called neurocomputers and neurochips, respectively [6]. While this new type of hardware is still not available commercially in significant amounts, it is always possible to simulate the parallel operation of a neural network in a sequential computer, by means of special algorithms and development systems (e.g., [7-9]). Several useful, up-to-date reviews of artificial neural networks concepts and applications have been written (e.g., [10, 11]), and the reader is referred to them for background material on the subject of this review. 2. Implications of ANNs for Biomedicine The emergence of connectionist models has brought forward important implications for the biomedical applications of Artificial Intelligence [12, 13]. There is, of course, a proximity between the field of study of connectionist systems and the Neurosciences, since many key concepts were drawn from investigations on the organization and function of biological neural networks, the neural basis of cognition and perception, the mechanisms used by the central nervous system to recognize visual and auditory patterns, etc. [14]. In reality, most of the artificial neural network models are only vaguely neuromorphic, being more of inventions than of faithful models of neural function. Therefore, its mechanisms of operation could be considered as more similar to classical algorithms and statistical methods of numerical taxonomy and pattern recognition, such as linear and nonlinear discriminant analysis, cluster analysis, etc. [15], or even to logic programming based on graphs, objects and networks [16]. Even so, it is clear that new mathematical and statistical techniques which were developed for the quantitative study of natural neural networks [17] and for modeling them numerically and statistically (e.g., [18-20]) offer a great potential for the enhancement of connectionist technology. On the other hand, connectionist models have been shown to be very useful and effective alternatives for implementing intelligent systems (application software and instruments) in several areas of Biology and Medicine, mainly those needing complex pattern recognition and classification (such as in the processing and understanding of signals and images), as well as decision support systems. In the following sections we will review the state of the art in each of these applications. The review doesn't intend to be exhaustive, since the published literature is already extensive and space restrictions impede a more profound treatment. However, a full review is available upon demand, from the author's laboratory [21]. Biological Signal Processing Artificial neural networks lend themselves very naturally to applications involving biological signal processing and recognition [22]. It has been proved that ANNs perform better and quickier than conventional methods in situations where noise and uncertainty is present, and that they can be used to implement several functions, such as filters, artifact detection, time-series and Fourier analysis, classification and recognition of complex patterns, etc. In electrocardiology, ANNs have been used to process and analyze the propagation of intracardiac signals [23], EKG signals [24], dynamic EKG (Holter analysis) [25] and phonocardiograms [26]. In clinical neurophysiology, ANNs have been applied to the analysis and classification of EMG [27] and EEG [28] signals, as well as brain evoked potentials [29], etc. Many of the techniques which were developed in this realm find pratical applications in decision support tasks in several areas of Medicine and Surgery, such as the estimation of the anesthetic level by a neural network which was trained to analyse the EEG spectral signature in each phase of the narcose [30], in the analysis of sleep patterns in adults [31] and children [32], and many others. Especially interesting is the fact that ANNs may help the study of biological neural networks, by means, for example, of the analysis of complex multi-unit electrical activity [19, 33], thus closing the circle of mutual influence between the Neurosciences and Connectionist Science. The development of instruments and techniques which are able to record simultaneously the activity of thousands of neurons (e.g., [34]) will put an enormous load on computational resources, which only ANNs will be able to meet efficiently and quickly. Another interesting possibility is the use of ANNs to control intelligent neurosensorial and motor prostheses [35, 36]. Medical and Biological Image Processing Due to the modular and repetitive structure of imaging systems, parallel processing has always been a successful approach to medical and biological signal processing. Thus, it is not surprising that ANNs have also enjoyed a great number of applications in this area, with several important advantages over conventional systems, particularly in the presence of noise and image variability. Promising applications have been demonstrated in areas as diverse as digital radiography [37], computed tomography [38], scintilography [39], SPECT [40] and PET [41] imaging, ultrasound [42] and microscopy [43]. A large variety of image processing techniques can be also implemented by means of ANNs, such as parameter selection [41], vector quantization [44], image compression [45], correction or scattering [46], edge detection [47, 48], image segmentation [49], tridimensional reconstruction [50], etc. Particulary impressive are many demonstrations of the capability of neural networks to classify and recognize complex and space- and orientation-variant medical images. ANNs perform equally well or better than non-connectionist approaches in areas such as scene [51] and texture [52] analysis, automatic identification and extraction of image features [53], recognition of biological patterns, such as chromossomes [54], tumors [55] and metastases [56]. For example, in a study by Silverman and Noetzel [42] a multiple ANN was successfuly trained to recognize and pinpoint the exact location of retinal tumors imaged by ultrasound, and later to classify them according to tumor type. Medical Decision-Support Systems The connectionist paradigm of Artificial Intelligence is very flexible, in the sense that it is able to accept and to process several kinds of formal representation of knowledge, such as feature vectors, probabilities, frame/hierarchical structures, semantic and belief networks, and even production rules. As general pattern classifiers, ANNs can be used as efficient tools in areas where traditional, heuristics-oriented, logical-based expert systems have dominated so far, such as medical diagnosis and therapy. ANNs are especially useful, in contrast to conventional AI approaches, in representing and treating uncertain, probabilistic, approximate, fuzzy or noisy concepts and data, either at the input or the output level. Thus, there has been an increase in the number of applications devoted to clinical decision-support using ANNs. Hripcsak [57] has analysed in detail the performance of ANNs in relation to statistical decision-making processes. ANN's applications as this have already found way in many medical decision support areas, as diverse as neurology [58], psychiatry [59], ophtalmology [60], endocrinology [61], cardiology [62], gastroenterology [63, 64], dermatology [65], neonatology[66], surgery [67], orthopedy [68], anesthesiology [69], oncology [70], radiology [71], radiation therapy [72], clinical pathology [73], nursing [74] and alternative medicine [75]. One of the criticisms which are often made about the usefulness of ANNs in the realm of clinical expert systems has been its "black-box" nature, i.e., its inability to explain logically its decisions. This is considered unacceptable in terms of user interface in the medical area. However, recent work has demonstrated that neural networks, when used to represent hierarchical and semantic structures, is able to explain its conclusions to users, as well [76]. Another interesting application of ANNs using non-supervised learning is the automatic discovery of complex patterns involving medical databases with hundreds of variables and thousands of records [77]. Other Applications There are many computation-intensive, ill-posed problems in Biology and Medicine which constitute ideal subjects for ANN application. The extensive development of computational techniques which is taking place in Molecular Biology and Genetics, for example, is already making use of ANNs to compare sequences of nucleic acids [78] and proteins [79], classification of chromossomes [80] and genes [81], analysis and prediction of secondary and tertiary structure of proteins [82, 83], etc. Another interesting field is the prediction of structure-activity relationships and physico-chemical properties [84]. 4. Conclusions The enormous increase in the literature on biological and medical applications of connectionist systems in the last three years offers strong support to the idea that this new paradigm in Artificial Intelligence is gaining strenght and eventually will be a major R & D field in itself. Although this increase could be traced to a general, unspecific enthusiasm following the introduction of a brand new methodology in any field, many serious investigations have provided evidence that a connectionist system is the preferred method to be used when it is required an adaptive solution to ill-defined and complex problems of pattern classification and recognition. Not only they are easier to implement and to train using examples, rather than complicated and unreliable heuristics, such as genetic algorithms or rule- modification schemes; but also they have the strong incentive that, once true massively-parallel neuromorphic computers are available, they will provide unprecedented, blinding-fast devices to implement intelligent applications in this field. There are some neural chips in the market, for example, which, when coupled to proper software, are able to achieve speeds in excess of two billion connections per second. It is predictable that there are some biomedical application areas where artificial neural networks will be the predominant solution within a few years. Such is the case of intelligent biomedical instruments, which will be able to implement easily and inexpensively some tasks that, today, require huge computing power, such as real- time classification of spectra and images, closed-loop control of automatic systems [85] and intelligent prosthetic devices [86]. The field of research on new models, topologies, algorithms and physical realization of ANNs is already thriving enormously, and eventually will discover how to circumvent may of its present limitations. It is foreseeable, also, that future intelligent systems will be, more and more, hybrid in nature, incorporating ANNs where they are more advantageous, to logics-based systems where high-level semantic and heuristic processing is valuable. The essential interdisciplinarity of Connectionist Science, where a large breadth of knowledge is required (mathematics, modeling, simulation, computer sciences, neurosciences, cognitive science, Artificial Intelligence) poses some important obstacles to its wider dissemination. However, this can be solved with proper training and teamwork [87]. References [1] McCulloch, W.; Pitts, W. - A logical calculus of the ideas immanent in nervous activity. Bull. Mathem. Biophys., 7: 115-133, 1943. [2] Rosenblatt, F. - The perceptron: a probabilistic model for information storage and organization in the brain. Psychol. Rev., 65: 368-408, 1958. [3] Kohonen, T. - Self Organization and Associative Memory. Berlin: Springer-Verlag, 1984. [4] Minsky, M.; Papert, S. - Perceptrons. Cambridge, MA: The MIT Press, 1969. [5] Eberhart, R.C.; Dobbins, R.W. - Early neural network development history: the age of Camelot. IEEE Engineer. Med. Biol. Magaz., 9(3):, 15-18, 1990. [6] Stubbs, D.F. - Neurocomputers. MD Computing, 5(3): 14-24, 1988. [7] Zipser, D.; Rabin, D. - P3: A parallel network simulating system. In: Rumelhart, D.E.; McClelland, J.L. (Eds.) - Parallel Distributed Processing: Explorations in the Microstructure of Cognition, Vol. 1. Cambridge, MA: The MIT Press, p.488-506, 1986. [8] D'Autrechy, C.L.; Reggia, J.A.; Sutton III, G.G.; Goodall, S.M.; Tagamets, M.A. - Developing connectionist models with MIRRORS/II. Proc. 12th. Ann. Symp. on Comput. Appl. Med. Care. New York: IEEE Press, p. 276-281, 1988. [9] Arruda-Botelho, A.G.; Sabbatini, R.M.E. - NEUROL: a high-level language and microcomputer program for the description and simulation of neural architectures. Res. VI Reun. An. Feder. Soc. Biol. S. Paulo (Brazil), 1991. [10] Simpson, P.K. - Artificial Neural Systems: Foundations, Paradigms, Applications and Implementations. Pergamon Press, London, 1990. [11] Nelson, M.M.; Illingworth, W.T. - A Practical Guide to Neural Nets. Addison Wesley, 1990. [12] Reggia, J.A. - Artificial neural systems in medical science and practice. MD Computing, 5(3): 4-6, 1988. [13] Reggia, J.A.; Sutton III, G.G. - Self-processing networks and their biomedical implications. Proc. IEEE, 76(6): 580-92, 1988. [14] Schreter, Z. - Connectionist models and their relation to AI, Psychology, and Neuroscience. Proc. II Int. Symp. Artif. Intell. Expert Syst., Vol. 2, p.187-210, 1988. [15] Huang, W.Y. & Lippman, R.P. - Comparisons between neural nets and conventional classifiers. Proc. 1st. International Neural Networks Conference. IEEE, New York, p. 485, 1987. [16] Feldman, J.A.; Fanty, M.A.; Goddard, N.H.; Lynne, K.J. - Computing with structured connectionist networks. Comm. ACM, 31(2): 170-187, 1988. [17] Achacoso, T.B.; Fernandez, V.; Nguyen, D.C.; Yamamoto, W.S. - Computer representation of the synaptic connectivity of Caenorhabditis elegans. Proc. 13th. Ann. Symp. Comput. Appl. Med. Care. New York: IEEE Press, p. 330-334, 1989. [18] Hartline, D.K. - Models for simulation of real neural networks. Neural Networks, 1 (Suppl.1): 256, 1988. [19] Tam, D.C.; Perkel, D.H.; Tucker, W.S. - Correlation of multiple neuronal spike trains using the back-propagation error correction algorithm. Neural Networks, 1 (Suppl.1): 277, 1988. [20] Wilson, M.A.; Bhalla, U.S.; Uhley, J.D.; Bower, J.S. - GENESIS: A system for simulating neural networks. In: Touretzky, D. (Ed.) - Advances in Neural Network Information Processing Syst.. San Mateo, CA: Morgan Kauffmann Publ., p.257 ff., 1989. [21] Sabbatini, R.M.E. - Applications of artificial neural networks in Biology and Medicine: a review. Biomedical Informatics Technical Reports, State University of Campinas, Campinas, Brazil, 1992. [22] Graupe, D.; Liu, R.W.; Moschytz, G.S. - Applications of neural networks to medical signal processing. Proc. 27th IEEE Conf. Decis. Contr. New York: IEEE Press, Vol.1, p. 343-7, 1988. [23] Dassen, W.R.; Mulleneers. R.G.; Den Dulk. K.; Smeets, J.R.; Cruz, F.; Penn, O.C.; Wellens, H.J. - An artificial neural network to localize atrioventricular accessory pathways in patients suffering from the Wolff-Parkinson-White syndrome. PACE 13(12 Pt 2):1792-6, 1990. [24] Linnebank, A.C.; Sippens-Groenenweg, A.; Grimbergen, C.A. - Artificial neural networks applied in multiple lead electrocardiography. Rapid quantitative classification of ventricular tachycardia QRS integral patterns. Proceed. 12th. Ann. Conf. IEEE Engineer. Med. Biol. Soc., Philadelphia, PA, USA, 1990. [25] Iwata, A.; Nagasaka, Y.; Suzumura, N. - Data compression of the ECG using neural network for digital Holter monitor. IEEE Engineer. Med. Biol. Magaz., 9(3): 53-57, 1990a. [26] Barschdorff, D.; Ester, S.; Dorsel, T.; Most, E. - A new phonographic technique for congenital and acquired heart disease using neural networks. Biomed. Tech. 35(11): 271-9, 1990. [27] Graupe, D.; Vern, B.; Gruener, G.; Field, A.; Huang, Q. - Decomposition of surface EMG signals into single fiber action potentials by means of neural networks. 1989 IEEE International Symposium on Circuits and Systems. New York, NY: IEEE, p. 1008-1011, Vol. 2, 1989. [28] Eberhart, R.C.; Dobbins, R.W.; Webber, W.R.S. - CASENET: a neural network tool for EEG waveform classification. In: Proc. II Ann. IEEE Symp. Comput.-Based Med. Syst. Washington, DC: IEEE Comput. Soc. Press, p. 60-8, 1989. [29] Bruha, I.; Madhavan, G.P. - Need for knowledge-based subsystem in evoked potential neural-net recognition system. In: Kim, Y.; Spelman, F.A. (Eds.) - Images of the Twenty First Century. Proceed. Ann. Internat. Conf. IEEE Engineer. Med. Biol. Soc., Seattle, WA Nov 1989. New York, NY: IEEE, Vol. 6, p. 2042-3, 1989b. [30] Watt, R.C.; Navabi, M.J.; Scipione, P.J.; Hameroff, S.R.; Maslana, E.S. - Neural network estimation of anesthetic level using EEG spectral signatures. Proceed. 12th. Ann. Conf. IEEE Engineer. Med. Biol. Soc., Philadelphia, PA, USA, 1990. [31] Schaltenbrandt, N.; Minot, R.; Mach, J.-P.; Lengell‚, R. - All- night sleep stage scoring using a neural network model. Proc. III Int. Worksh. Neural Networks & Their Applications (NEURO-NIMES 90). Nimes, Nov. 1990. [32] Pfurtscheller, G.; Litscher, G. - Analysis of sleep patterns in babies using neural networks - preliminary results. Medical Informatics Europe MIE'91. Berlin: Springer-Verlag, 1991. [33] Iezzi, R.; Micheli-Tzanakou, E. - Neural network analysys of neuronal spike-trains. Proceed. 12th. Ann. Conf. IEEE Engineer. Med. Biol. Soc., Philadelphia, PA, USA, 1990. [34] Nash, P.L.; Muljadi, P.; Wayner, M.J.; Senseman, D.J. - High- speed imaging of electrical activity: watching the brain think on MTV/2. Neural Networks, 1 (Suppl.1): 268, 1988. [35] Uth, J.; Graupe, D. - Neural networks for function discrimination in EMG controlled FES walking for paraplegics. Neural Networks, 1 (Suppl.1): 469, 1988. [36] Liu, R.; Tong, L.; Deng, I. - An on-line learning neural network for discrimination of walk functions in paraplegics. 1989 IEEE Symp. on Circuits and Systems. New York, NY: IEEE, p. 1017-1020, Vol.2, 1989. [37] Gross, G.W.; Boone, J.M.; Greco-Hunt. V.; Greenberg, B. - Neural networks in radiologic diagnosis. II. Interpretation of neonatal chest radiographs. Investigative Radiology, 25(9): 1012-1016, 1990. [38] Fredrick, B. deB. - Implementation and evaluation of a neural network algorithm for magnetic resonance image segmentation. Proc. 4th SPIE Conf. Med. Imaging. SPIE, Bellingham, WA, USA, 1990. [39] Anthony, D.; Hines, E.; Taylor, D.; Barham, J. - An investigation into the use of neural networks for an expert system in nuclear medicine image analysis. Proc. 3rd Int. Conf. Image Process. and Its Applic. London, IEE, p. 338-342, 1989. [40] Tourassi, G.D.; Floyd Jr., C.E.; Bowsher, J.E.; Munley, M.T.; Coleman, R.E. - Application of neural networks for lesion detection in SPECT. Proceed. IEEE Nuclear Science Symposium and Medical Imaging Conference. Santa Fe, New Mexico, USA, Nov. 1991. [41] Miller, L.F.; Smith, G.T.; Wu, Y.; Uhrig, R.E. - Evaluation of neural networks for parameter identification from positron emission tomography scans. Trans. Am. Nucl. Soc. 62: 5-6, 1990. [42] Silveman, R.H.; Noetzel, A.S. - Image processing and pattern recognition in ultrasonograms by backpropagation. Neural Networks, 3(3): 593-604, 1990. [43] Dytch, H.E.; Wied, G.L.; Bibbo, M. - Neural nets as tools for expert systems in objective histopathology. Proc. Ann. Int. Conf. of IEEE Eng. Med. Biol. Soc. New York: IEEE, p.1375-6, 1988. [44] He, Y.; Zhang, Q.; Ye, Y.; Li, Z. - Vector quantization by using neural networks. Proc. 4th SPIE Conf. Med. Imaging. SPIE, Bellingham, WA, USA, 1990. [45] Anthony, D.; Hines, E.; Taylor, D.; Barham, J. - A study of data compression using neural networks and principal components analysis of pulmonary scintigrams. IEE Colloquium on Biomedical Applications of Digital Signal Processing. London, UK: IEE, p. 271-5, 1989. [46] Boone, M.; Seibert, J.A. - Neural network scatter correction technique for digital radiography. Proc. 4th SPIE Conf. Med. Imaging. SPIE, Bellingham, WA, USA, 1990. [47] Asada, N.; Eiho, S.; Doi, K.; MacMahon, H.; Montner, S.M.; Giger, M.L. - Pilot study of image segmentation and decision making by neural network. Proc. 4h SPIE Conf. Med. Imaging. SPIE, Bellingham, WA, USA, 1990. [48] Sabbatini, R.M.E. - A computer program to perform edge extraction in computed tomography using neural networks . An. I Congr. Latinoamer. Inform. Salud, Habana, Cuba, 1992. [49] V zquez, R.; Medina, F.I. - Use of artificial neural networks for X-ray image segmentation. Proceed. IEEE Nuclear Science Symposium and Medical Imaging Conference. Santa Fe, New Mexico, USA, Nov. 1991. [50] Venaille, C.; Mischler, D.; Collorec, R.; Catros, J.Y.; Coatrieux, J.L. - Automatic 3-D reconstruction of vascular networks from three projections: s simulated annealing approach. In: Kim, Y.; Spelman, F.A. (Eds.) - Images of the Twenty-First Century. Proceed. Ann. Int. Conf. IEEE Engineer. Biol. Med. Soc. New York: IEEE Press, Vol. 2, p. 565-6, 1989. [51] Koutsougeras, C.; Barad, H.; Martinez, A. - Neural networks in scene analysis. Proceed. Applications of Artificial Neural Networks. Bellingham, WA, USA: SPIE, Vol. 1294, p. 86-93, 1990. [52] Parikh, J.; DaPonte, J.S. - Application of neural networks to pattern recognition problems in remote sensing and medical imagery. In: Rogers, S.K., (Ed.) Proceed. Conf. Applications of Artificial Neural Networks. Bellingham, WA: SPIE, Vol. 1294, p. 146-60, 1990. [53] Lo, S.C.; Freedman, M.; Liu, F.; Krasner, B.; Mun, S.K. - Computer-assisted diagnosis for lung nodule detection using a neural network technique. Proceed VI Medical Imaging Conference. Bellingham, WA: SPIE, 1992. [54] Takaya, K.; Sato, K. - Chromosome classification using associative memory techniques. Proc. 3rd World Conf. on Med. Inform. Amsterdam: North Holland, p.222-225, 1980. [55] Dawson, A.E.; Austin, R.E.; Weinberg, D.S. - Nuclear grading of breast carcinoma by image analysis - classification by multivariate and neural network analysis. Amer. J. Clin. Path., 95 (4): S29-S37, 1991. [56] DaPonte, J.S.; Parikh, J.; Katz, D.A. - Detection of liver metastasis using the backpropagation algorithm and linear discriminant analysis. Proceed. SPIE Conf. Applications of Artificial Intelligence and Neural Networks, Bellingham, WA: SPIE, 1991. [57] Hripcsak, G. - Using connectionist modules for decision support. Meth. Inform. Medicine, 29: 167-181, 1990. [58] Apolloni, B,; Avanzini, G.; Cesa-Bianchi, N.; Ronchini, G. - Diagnosis of epilepsy via back-propagation. Proceed. IJCNN-90 - International Joint Conference on Neural Networks, Washington, DC., p. 571-574, Vol. 2, 1990. [59] Mulsant, B.H.; Servan-Schreiber, E. - A connectionist approach to the study of dementia. Proc. 12th. Ann. Symp. Comput. Appl. Med. Care. New York: IEEE Press, p. 245-250, 1988. [60] Coffey, D.; Banks, G. - A connectionist visual field analyzer. Proc. 13th. Ann. Symp. Comput. Appl. Med. Care. New York: IEEE Press, p. 276-282, 1989. [61] Smith, J.W.; Everhart, J.E.; Dickson, W.C.; Knowler, W.C.; Johannes, R.S. - Using the ADAP learning algorithm to forecast the onset of diabetes mellitus. Proc. 12th. Ann. Symp. Comput. Appl. Med. Care. New York: IEEE Press, p. 261-265, 1988. [62] Harrison, R.F.; Marshall, S.J.; Kennedy, R.L. - A connectionist aid to the early diagnosis of myocardial infarction. Proceed. Third European Conf. Artificial Intelligence in Medicine, Maastricht, The Netherlands, June 1991. [63] Yoshida, K.; Hayashi, Y.; Imura, A. - A connectionist expert system for diagnosing hepatobiliary disorders. Proc. 6th World Conf. Med. Inform. Amsterdam: North Holland, p. 116-120, 1989. [64] Eberhart, R.C.; Dobbins, R.W. - Neural networks versus bayesian diagnosis of appendicitis. Proceed. 12th. Ann. Conf. IEEE Engineer. Med. Biol. Soc., Philadelphia, PA, USA, 1990. [65] Yoon, Y.; Peterson, L.L.; Bergstresser, P.R. - DESKNET: The Dermatology expert system with knowledge-based network. Neural Networks, 1 (Suppl.1): 477, 1988. [66] McMillan, M.M.; Walter, D.C. - Automated Medical Student - a computational model of skill acquisition and expert performance. Proc. Sec. Ann. IEEE Symp. Comput.-Based Med. Syst.. Washington, DC: IEEE Computer Soc. Press. p. 108-13, 1989. [67] Pham, K.M.; Degoulet, P. - MOSAIC: medical knowledge processing based on a macro-connectionist approach to neural networks. Proc. 6th World Conf. Med. Inform. Amsterdam: North Holland, p.82-86, 1989. [68] Bounds, D.G.; Lloyd, P.J.; Mathew, B.G. - A comparison of neural network and other pattern recognition approaches to the diagnosis of low back disorders. Neural Networks, 3(3): 583-592, 1990. [69] Garcia, O.R.; Celso de Lima, W.; Duarte, D.F. - Expert system for evaluation and proposal of anesthetic procedures with neural networks. Proceed. Intern. Conf. Systems Engineer., Las Vegas, Nevada, July 1990. [70] Hudson, D.L.; Cohen, M.E.; Anderson, M.F. - Determination of testing efficacy in carcinoma of the lung using a neural network model. Proc. 12th. Ann. Symp. Comput. Appl. Med. Care. New York: IEEE Press, p. 252-255, 1988. [71] Boone, J.M.; Sigillito, V.G.; Shaber, G.S. - Neural networks in Radiology - An introduction and evaluation in a signal detection task. Med. Phys., 17(2):234-241, 1990. [72] Morrill, S.M.; Lane, R.G.; Rosen, I.I. - Constrained simulated annealing for optimized radiation treatment planning. Comput. Progr. Meth. Biomed. 33(3): 135-44, 1990. [73] F”rsstrom, J.; Fogstr”m, M. - Using connectionist approach for finding ideal cut-off values for quantitative laboratory tests). Medical Informatics Europe MIE'91. Berlin: Springer-Verlag, 1991. [74] Harvey, R.M. - A neural network as a potential means for aiding nursing diagnosis. Proceed. 14th. Annual Symposium on Computer Applications in Medical Care. New York: IEEE Press, p. , 1990. [75] Chen, W.; Wang, Z.; Guo, R. - NNTCM - A neural network applied to traditional Chinese medical diagnosis. Proc. 6th World Conf. Med. Inform. Amsterdam: North Holland, p. 92-95, 1989. [76] Diederich, J. - An explanation component for a connectionist inference system. In: Aiello, L.C. (Ed.) - Proceed. 9th. Europ. Conf. Artif. Intellig., ECAI 90, Stockholm, Sweden. London, UK: Pitman, p. 222-7, 1990. [77] Irani, E.A.; Long, J.M.; Slagle, J.R. - Experimenting with artificial neural networks. Proc. Symp. Eng. Comput.-Based Med. Syst.. Washington, DC: IEEE Computer Society Press, p. 45-46, 1988. [78] Furutani, H.; Yamamoto, K.; Kitazoe, Y.; Ogura, H. - Analysis of thalassemia beta-globin genes by neural networks: prediction of abnormal splicing. Proc. 6th World Conf. Med. Inform. Amsterdam: North Holland, p. 96-100, 1989. [79] Bengio, Y.; Pouliot, Y. - Efficient recognition of immunoglobulin domains from amino acid sequences using a neural network. Comput. Appl. Biosci., 6(4): 319-24, 1990. [80] Noehte, S.; Manner, R.; Hausmann, M.; Horner, H.; Cremer, C. - Classification of normal and aberrant chromossomes by an optical neural network in flow citometry. Optical Computing 1989. Washington, DC: Optical Soc. America, AOSR/DARPA Technical Digest Series, Vol. 9, p. 14-17, 1989. [81] Wu, C.H.; Whitson, G.M. - Artificial neural system for gene classification using a domain database. Proc. 1990 Comput. Sci. Conf.. New York, Association of Computing Machinery, 1990. [82] Quian, N.; Sejnowski, T. - Predicting the secondary structure of globular proteins using neural network models. J. Mol. Biol., 202: 865-884, 1988. [83] Bohr, H.; Bohr, J.; Brunak, S.; Cotterill, R.M.J.; Fredholm, H.; Lautrup, B.; Petersen, S.B. - A novel approach to prediction of the 3-dimensional structures of protein backbones by neural networks. FEBS Letters, 261(1):43-46, 1990. [84] Aoyama, T; Suzuki Y; Ichikawa H - Neural networks applied to structure activity relationships. J. Med. Chem., 33(3): 905-908, 1990. [85] Werbos, P.J. - Neural networks for control: an overview. Proceed. 12th. Ann. Conf. IEEE Engineer. Med. Biol. Soc., Philadelphia, PA, USA, 1990. [86] Khoshaba, T.; Badie, K.; Hashemi, R.M. - EMG patterns classification based on backpropagation neural network for prosthesis control. Proceed. 12th. Ann. Conf. IEEE Engineer. Med. Biol. Soc., Philadelphia, PA, USA, 1990. [87] Sabbatini, R.M.E. - NEURONET: a shell connexionist system for automatic pattern classification in Biomedicine. An. I Forum Ciˆncia e Tecnologia em Sa£de. S. Paulo, Brazil, 1992. (c) Copyright 1992 Renato Marcos Endrizzi Sabbatini Published in: K.C. Lun, Degoulet, P. & Piemme, T. (Eds.) Proceedings of the 7th International Congress on Medical Informatics. Geneva, Switzerland, September 1992. Amsterdam: North-Holland, 1992, p. 418-425. ADRESS FOR CORRESPONDENCE: Renato M.E. Sabbatini, PhD Center for Biomedical Informatics State University of Campinas P.O. Box 6005 Campinas, Sao Paulo, 13081-970, Brazil Tel. +55 192 39-7130 Fax. +55 192 39-4717 Tlx. +55 19 1150 uec br