Seeking an optimal approach for Computer-aided Diagnosis of Pulmonary Embolism

Nahid Ul Islam, Zongwei Zhou, Shiv Gehlot, Michael B. Gotway, Jianming Liang

Abstract: Highlights•We benchmark supervised and self-supervised ImageNet models for PE classification task.•We propose an Embedding-based ViT (E-ViT) framework for PE exam-level diagnosis.•E-ViT exploits class-level and exam-level embedding, resulting in a significant performance gain.•E-ViT can process varying slice numbers, allowing greater flexibility.•Vessel-oriented representation enhances performance with increased image dimensions.