Fuk Hay Tang PhD, Presenter: Nothing to Disclose

   Our finding suggested that our CAD scheme could improve diagnosis accuracy of PE. We have found that intravascular filling defect was a reliable criterion for diagnosis of PE. The mean area of ROC curve (Az) of front-line doctors without and with CAD was 0.87 and 0.94 respectively. The performance of front-line doctors was improved significantly when the scheme was used (p<0.01). Among other diagnostic parameters, intravascular filling defect was significantly differenent for patients with and without PE.

  Ventilation-perfusion (VQ) scan was the major diagnostic tool  for pulmonary embolism  in imaging. Recently, computer tomography (CT) is more commonly used for diagnosis as it is relatively accurate due to multi-slice technology. However CT angiography diagnosis is time consuming, as it involves reading of hundreds of slices of images. It follows that it was highly desirable to use computer aided method in detecting and characterizing emboli to improve the efficiency of clinical management of such cases.

Our result showed that CAD scheme helped improve diagnostic performance as second opinion. Our result was comparable to the study by Shiraishi et al. (2003), where radiologists performed better with the CAD scheme.Tourassi et al. (1997) had performed similiar study using ventilation-perfusion lung scans and chest radiographs. Their results werecomparable to our study with Az value of that with ANN and that without ANN (physicians alone) was 0.91 and 0.81 respectively whereas that of our study was 0.95 and 0.81 respectively.

A retrospective study was performed for computed tomographic pulmonary angiograms requested for suspicious pulmonary embolism. Totally 40 image sets, 20 with PE and 20 normal, were selected. We proposed a CAD scheme implemented with 14 inputs and 2 outputs in predicting the presence or absence of PE. The 14 inputs were decided by the diagnostic parameters for PE through agreement with radiology experts. The CAD scheme was trained and tested using a round-robin method.Receiver operating characteristic (ROC) analysis was used to evaluate the diagnostic performance of physicians with and without using CAD scheme.

Tang, F, An Intelligent Computer Model for Improving Diagnostic Accuracy of Pulmonary Embolism for Computed Tomography Pulmonary Angiography.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12022991.html