Validating expert system rule confidences using data mining of myocardial perfusion SPECT databases

The authors' goal with this study was to use data mining techniques, applied to imaging and textual patient databases, to validate the confidences (certainty factors) of the heuristic rules in the authors' previously described Expert System, PERFEX/sup TM/. A relational database combining textual and imaging information was generated from 655 patients who had undergone both stress/rest myocardial perfusion SPECT and coronary angiography. Initial data mining was concentrated on heuristic rules involving myocardial perfusion defects and the LAD vascular territory. The results show the robustness of the expert system, and furthermore show that data mining of large databases combining textual and imaging information can be used to validate and potentially improve the confidence levels associated with heuristic rules in expert systems.