We introduce new techniques for medical data mining for the improved diagnosis in two medical domains.
For the domain of breast cancer diagnosis, we propose two approaches to test selection, to enable the selection of an optimal imaging modalty, one based on information maximization and one on subgroup discovery. For the domain of Alzheimer’s disease we focus on the correlation of image and non-image data. We start with clustering PET scans of patients to form groups sharing similar features in brain metabolism, we then explain the clusters by relating them to non-image variables, using an algorithm for relational subgroup discovery. In an alternative approach, we determine frequent itemsets on non-image data to describe possible image clusters, which then have to be combined into an overall clustering. This combination can be controlled by a variety of user-defined constraints, which we solve by integer linear programs.
«
We introduce new techniques for medical data mining for the improved diagnosis in two medical domains.
For the domain of breast cancer diagnosis, we propose two approaches to test selection, to enable the selection of an optimal imaging modalty, one based on information maximization and one on subgroup discovery. For the domain of Alzheimer’s disease we focus on the correlation of image and non-image data. We start with clustering PET scans of patients to form groups sharing similar features in...
»