Anatomically aware dual-hop learning for pulmonary embolism detection in CT pulmonary angiograms

• Published in: Computers in biology and medicine Vol. 174; p. 108464
• Main Authors: Condrea, Florin, Rapaka, Saikiran, Itu, Lucian, Sharma, Puneet, Sperl, Jonathan, Ali, A. Mohamed, Leordeanu, Marius
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01-05-2024; Elsevier Limited
• Subjects: Anatomically aware medical image recognition; Angiography; Artificial neural networks; Biology; Computed tomography; Computed Tomography Angiography - methods; Computer vision; Computers; CT pulmonary angiography; Datasets; Deep Learning; Deep neural networks; Diagnosis; Dual-hop learning; Embolism; Embolisms; Humans; Machine learning; Medical image analysis; Medical imaging; Neural networks; Neural Networks, Computer; Pulmonary arteries; Pulmonary Embolism - diagnostic imaging; Pulmonary embolism detection; Pulmonary embolisms; Workloads; Computer vision; Pulmonary embolism detection; Deep neural networks; CT pulmonary angiography; Medical image analysis; Dual-hop learning; Anatomically aware medical image recognition
• ISSN: 0010-4825; 1879-0534
• DOI: 10.1016/j.compbiomed.2024.108464

Pulmonary Embolisms (PE) represent a leading cause of cardiovascular death. While medical imaging, through computed tomographic pulmonary angiography (CTPA), represents the gold standard for PE diagnosis, it is still susceptible to misdiagnosis or significant diagnosis delays, which may be fatal for critical cases. Despite the recently demonstrated power of deep learning to bring a significant boost in performance in a wide range of medical imaging tasks, there are still very few published researches on automatic pulmonary embolism detection. Herein we introduce a deep learning based approach, which efficiently combines computer vision and deep neural networks for pulmonary embolism detection in CTPA. Our method brings novel contributions along three orthogonal axes: (1) automatic detection of anatomical structures; (2) anatomical aware pretraining, and (3) a dual-hop deep neural net for PE detection. We obtain state-of-the-art results on the publicly available multicenter large-scale RSNA dataset.
•Pulmonary embolism detection strong baseline in CTPAs using machine learning models.•Developing three anatomically inspired orthogonal directions of improvement.•State of the art results on recently released large scale dataset RSPECT.