Pulmonary Embolism Classification Using CT Images (and JMP Pro!)
Marie Guadard, PhD, Consultant – North Haven Group
This talk explores some of the new advanced predictive features of JMP Pro Version 9 in the context of a medical data mining application. Bootstrap forests and neural nets, as well as other techniques and the use of JMP Scripting Language, are employed in an effort to identify pulmonary embolisms using three-dimensional computed tomography (CT) data. The data we use for our training and validation sets formed the basis for the 2006 KDD (Knowledge Discovery and Data Mining) Cup competition. These data present interesting challenges: sparse and noisy data, multiple regions associated with a single pulmonary embolism, a spatial structure within patients, and non-traditional measures of sensitivity. We develop predictive models using the training data, choose a final model, and apply that model to the separate test set in order to assess its performance.