Early Pulmonary Embolism Detection from Computed Tomography Pulmonary Angiography Using Convolutional Neural Networks

In this study., we developed the first computer-aided detection (CAD) system aimed at triage patients with pulmonary embolism (PE) to reduce the death rate during the waiting period. Computed tomography pulmonary angiography (CTPA) is used for definite diagnosis of PE., and CTPA imaging reports are read by radiologists who suggest further management., which requires time and hence a waiting period to obtain a diagnosis. Patients may die during this waiting period., and a CAD method can triage patients with PE from those without PE. In this study., we proposed a CAD system to achieve the aforementioned purpose. Our purpose is different from related studies and CAD systems that were aimed at identifying key PE lesion images in images of patients with PE to expedite PE diagnosis. Our CAD system consists of a novel classification-model ensemble for PE detection and a segmentation model to label PE lesion on each image. We utilized data from the National Cheng Kung University Hospital and open resource to construct models. In the classification model., the algorithm achieved an area under the receiver operating characteristic curve of 0.88 (accuracy = 0.85). In the segmentation model., the mean intersection over union was 0.689. Overall., our CAD system successfully distinguished patients with PE from those without PE and automatically labeled the PE lesion to expedite PE diagnosis. Contribution-This is the first CAD system aimed at triage patients with PE that uses the multiple convolutional neural network architecture.