Data mining applications to dermatological diseases, colorectal cancer, diabetes, cardiology and medical audit systems

Abstract

Diagnosis of Dermatological Diseases This project aims to develop tools that would provide doctors with the ability to discover the typical characteristics of certain inflammatory dermatological diseases. It is hoped that this will aid doctors in accurately diagnosing the dermatological disorder, especially between two skin diseases with very similar characteristics such as psoriasis and atopic dermatitis. Prognostic Models for Colorectal Cancer Patients The goal of this project is to build a robust prognostic model that will predict length of survival for patients with colorectal cancer. Techniques employed for this purpose include Neural Networks, Case-based reasoning (CBR), Cox's Regression and Regression Tree Induction. Accomplishments have been the development of point estimates of survival from Cox's regression to allow direct comparison of Artificial Intelligence techniques with Cox's results and the development of techniques for building more perspicuous models that are intuitive to medical practitioners. Pre-empting Complications in Diabetic Patients Data, provided by Data Retrieval in General Practice, is currently being used to discover common trends that appear among diabetic patients. These trends can then be used to predict complications in a diabetic's life cycle. This will allow patients who are not at risk to be filtered out and those who are at risk to be carefully monitored. Data mining techniques such as Sequential Pattern Discovery are being employed for this purpose. Risk Assessment in Cardiology This project has been investigating the use of case-based reasoning and the incorporation of fuzzy logic to provide insight into the understanding of the underlying mechanisms of coronary heart disease. By identifying patients at risk early on, it is easier to stop the progression of this disease. A series of qualitative and quantitative information was gathered from middle aged men, such as weight, blood pressure, body fat and family medical histories over a three year time span, to help identify which factors contribute to coronary heart disease. Integrating Data Mining into Medical Audit Systems The aim of this project is to bring Data Mining into mainstream medical audit by integrating it with Medical statistics techniques already in use. This will encourage the practice of Evidence-based medicine, that is becoming a norm in western societies, providing IT support for it. The motivation behind this project is the realization that current Medical statistics techniques fall short of certain objectives and requirements of medical audit.