Data warehouse for machine learning: application to breast cancer diagnosis

Abstract

The diagnosis and treatment of breast cancer are very much studied in the field of medical research. Indeed, the early detection of this pathology allows for the best possible treatment and minimization of the after-effects.The business process "Diagnosis of breast cancer in women" is perfectly mastered by the practitioner and the medical profession and is widely documented in the literature. We propose a vision of “analysis of the existing” which allows to discover this process to the computer engineer in charge of the implementation of a system for aid in diagnosis and prognosis of the disease. This step is indispensable to discover the data generated during the business process and to discover in a later step all the characteristics that govern the disease. We define a system of diagnosis and prognosis of the disease Diaprog which uses the data of the classified and archived care records. We say that we use the historical data generated during the main process. We distinguish for this purpose between a production database and a data warehouse. The latter is a database fed from the various production databases accessible and made available in our system. In the case of our work, we call this data warehouse DDA (Diagnosis Data Archive). The data warehouse thus defined is a component of the medical diagnosis aid system Diaprog whose functional architecture also implements a learning module that allows the discovery of relationships between the characteristics that govern the disease.