Department of Electrical Engineering
University of Washington

Learning Patterns for Computer-Aided Diagnosis

Sponsors

Problem Statement and Objectives

Pap smears are widely used in North America, Europe, and other developed countries for early detection of cervical cancer. There are 50 million Pap smears analyzed annually in the U.S. alone. NeoPath, Inc. of Redmond, Washington sells or leases the AutoPap 300 System, an automated cytology analysis system. The current product is capable of automatically analyzing pap smears and classifying them as normal or as containing cancerous cells. NeoPath would like to move into more advanced tasks in cervical cancer management, starting with the task of computer-aided diagnosis to improve the quality and consistency of Pap smear diagnosis and treatment triage. Insteady of merely classifying cells as normal or abnormal, a computer-aided diagnosis system will aid cytologists in making slide diagnostic decisions and slide gradings. This will require computerized decision procedures with higher sensitivity and higher specificity than NeoPath's current products. For this reason, NeoPath would like to incorporate advanced learning techniques to create high-quality products while decreasing development time and costs.

Our project is to develop a pattern representation and associated learning techniques that can be used in NeoPath's current and future applications. The representation will be general and powerful enough to handle a wide variety of pattern recognition tasks. It will be designed to be an integral part of a supervised learning system, so that new patterns can be learned from real training data. The new technology will be developed to become an upgrade to the current AutoPap systems to add the computer-aided diagnosis capability. In addition, it could be used in future products as NeoPath develops new application areas.