A two step workflow for pulmonary embolism detection using deep learning and feature extraction

Pulmonary embolism is among the leading causes of death all over the world. In order to achieve a fast diagnosis and response from the medical team, a CT exam is used as a means to detect embolisms. Over the years, different methods of computer-aided diagnosis systems (CADs) were implemented to facilitate the analysis of a CT exam for pulmonary embolism detection, and due to the high amount of data produced in these exams, methods that use deep learning are also growing as of late. This paper proposes a pipeline to detect pulmonary embolisms from a CT exam. It uses a U-net network to detect embolisms candidates and classifies them between false positives and true positives using machine learning algorithms. The method achieved a dice score of 0.81 and an IoU of 0.79.