At the University of Calabria in Italy, the medical decision making process has been computerized. Physicians at the Cosenza General Hospital currently are using the diagnostic decision support system to help them with the timely identification of breast cancer in patients through the application of a well-defined set of classification data. Dr. Mimmo Conforti presented the system before the ITIS-ITAB'99 audience. As a mathematician, he explained the architecture from this particular point of view, emphasizing the powerful efficiency and effectiveness of Mathematical Programming approaches as the basic tools for the design of the CAMD or Computer Aided Medical Diagnosis system.
Mathematical programming models form ideal tools to automate the process of inducing a diagnosis from a small amount of clinical data and from the clinical experience of the medical expert. The classification phase constitutes the core of the medical diagnosis. All the signs or symptoms for identification of a disease are collected and introduced into a database, that is used for classification of the data. The features of the specific pathology are measured. These data measurements are assessed on the basis of the physician's clinical experience, which results in a knowlegde base or training set. A correct diagnosis can be drawn from accurate pattern recognition and classification.
Classification is the decision of assigning an object to a suitable class, based on a set of features, describing this object. In turn, the aim is to recognize a specific category in a set of data or predict a class membership. This kind of discriminating analysis depends on statistical approach, decision trees, rule-based reasoning, and neural networks to solve classification problems. Dr. Conforti however proposes a completely different approach, based on the use of mathematical programming methods. Discrimination between two sets of objects which in reality are not strictly separable calls for a compromise, that consists in using a linear programming formulation. The linear function that is subject to the linear constraints must be minimized as the average sum of violations of the misclassified objects.
This solution for the classification problem can easily be applied for medical diagnostic decision support systems since adequate algorithms and software are available. The CAMD prototype system aids the physician to discriminate between benign and malignant cells for the early detection of breast cancer. CAMD is composed of image processing tools based on cellular morphometry and an automatic classifier based on the mathematical programming tool. It uses image input from a microscope. A Web facility has been integrated into the system to allow remote diagnosis from any site by means of java-applets. The system is under current assessment and being validated in collaboration with the Central Pathology Department at Cosenza General Hospital.
In the future, the CAMD team will try to improve application of mathematical programming models, using some new parallel algorithms. The work is being performed in interaction with the end-user institutions. The extension of the system to other types of disease is under investigation at present. For further details, we refer to the home page of the CAMD project, as well as to the VMW article on The Web-integration of a Computer Aided Medical Diagnosis tool for breast cancer.