Interpreting the results of myocardial perfusion scintigraphy by machine learning

The diagnostic accuracy of stress myocardial perfusion scintigraphy (MPS) is well defined, being for planar studies around 85% and for SPECT 90% subsequently. There are reports on experiments of interpretation MPS results by Machine Learning (ML) methods. Applied data in these experiments were mainly the scintigraphyc images, in some reports stress and/or laboratory data also. In our experiments we tested the ML program Naive Bayesian Classifier to predict the persistency of CAD from the planar MPS results in a group of 327 pts. As ML methods can simultaneously deal with whatever number of data, we included in our experiments also the results of history/status/lab/ECG and exercise ECG for each pt. The result was than compared with the standard MPS reading ones. In standard results the sensitivity was 85%, specificity 86% and accuracy 85%. In ML results sensitivity was 92%, specificity 88% and accuracy 91, 5%. ML has in this way proven to be an useful tool in MPS results interperatation.