2021/07/13 by Sudhir Suman, Gagandeep Singh, Suman, Sudhir +13
Medicine · #Acute Ischemic Stroke Management #Cardiac Imaging and Diagnostics #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #Venous Thromboembolism Diagnosis and Management #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2107.06276
openalex publication_date 2021/07/13 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
With more than 60,000 deaths annually in the United States, Pulmonary\nEmbolism (PE) is among the most fatal cardiovascular diseases. It is caused by\nan artery blockage in the lung; confirming its presence is time-consuming and\nis prone to over-diagnosis. The utilization of automated PE detection systems\nis critical for diagnostic accuracy and efficiency. In this study we propose a\ntwo-stage attention-based CNN-LSTM network for predicting PE, its associated\ntype (chronic, acute) and corresponding location (leftsided, rightsided or\ncentral) on computed tomography (CT) examinations. We trained our model on the\nlargest available public Computed Tomography Pulmonary Angiogram PE dataset\n(RSNA-STR Pulmonary Embolism CT (RSPECT) Dataset, N=7279 CT studies) and tested\nit on an in-house curated dataset of N=106 studies. Our framework mirrors the\nradiologic diagnostic process via a multi-slice approach so that the accuracy\nand pathologic sequela of true pulmonary emboli may be meticulously assessed,\nenabling physicians to better appraise the morbidity of a PE when present. Our\nproposed method outperformed a baseline CNN classifier and a single-stage\nCNN-LSTM network, achieving an AUC of 0.95 on the test set for detecting the\npresence of PE in the study.\n