Adaptive Neural Network Imaging in Medical Systems Constantinos S. Pattichis lT2, Marios S. Pattichis ‘Department of Computer Science, University of Cyprus, Kallipoleos 75, Department of Electrical and Computer Engineering, University of New Mexico, Albuquerque, NM 87131-1356, USAemail: pattichis@eece.unm.edu P.O. Box 20537, CY 1678 Nicosia, Cyprus email: pattichi@ucy.ac.cy 2 Abstract Recent technological advances in medicine facilitated the development of sophisticated equipment enabling the better delivery of health care services. In parallel, artificial neural networks emerged as promising tools for the application and implementation of intelligent systems. The aim of this paper is to provide a snapshot of the application of neural network systems in medical imaging. The paper will highlight neural network applications in the analysis of cervicovaginal smears, mammography, microscopy, ultrasound imaging, and lesion placement in pallidotomy. It is anticipated that the application of neural network systems in medicine will provide the framework for the development of emerging medical systems, enabling the better delivery of health care. I. Introduction The overall objective of Computer Aided Diagnostic (CAD) systems is to enable the early diagnosis, disease monitoring, and better treatment. The advantages of CAD systems can be summarized as follows: Standardization. Diagnoses obtained from different laboratories using similar criteria can be verified. Sensitivity. Findings on a particular subject may be compared with a database of normal values and/or a decision can be made by a CAD system deciding whether or not an abnormality exists. Specificity. Findings may be compared with databases for various diseases and/or a decision can be made by the CAD system with respect to the type of abnormality. Equivalence. Results from a series of examinations of the same patient may be compared to decide whether there is evidence of disease progression or of response to treatment. In addition, the findings of different CAD systems can be compared to determine which are more sensitive and specific. Eficacy. The results of different treatments can be more properly evaluated. Medical imaging provides vital information for CAD systems. 0-7803-7147-X/01/$10.0002001 IEEE The objective of this paper is to present a snapshot of adaptive neural network applications in medical imaging and how these techniques can be integrated in CAD systems. According to Widrow and Stearns [l], the major characteristics of an adaptive system can be summarized as follows: 1. They can automatically adapt (self- optimize) in the face of changing (nonstationary) environments and changing system requirements. 2. They can be trained to perform specific filtering and decision-making tasks. 3. They can extrapolate a model of behavior to deal with new situations after having been trained on a finite and often small number of training signals or patterns. 4. To a limited extent, they can repair themselves; that is, they can adapt around certain kinds of internal defects. 5. They can usually be described as nonlinear systems with time-varying parameters. 6. Usually, they are more complex and difficult to analyze than non-adaptive systems, but they offer the possibility of substantially increased system performance when input signal characteristics are unknown or time varying. Linked with adaptive systems, are artificial neural networks or artificial neural network systems. According to Haykin [2], [3], a neural network can be defined as follows: “A neural network consist of the interconnection of a large number of nonlinear processing units called neurons; that is, the nonlinearity is distributed throughout the network. We are interested in a particular class of neural networks that Zeam about their environment in a supervised manner. We have a desired response that provides a target signal, which the neural network tries to approximate during the learning process - achieved by adapting synaptic weights, in a systematic manner.”In the context of adaptive signal processing applications, neural networks offer the following advantages [2], [3]: nonlinearity, input-output mapping, weak statistical assumptions, learning capability, generalization, fault tolerance, and VLSI implementation. In this paper, a very brief review of adaptive neural network applications in medical imaging is given. In the next section, the results of literature search on neural 313