Cardiac-CLIP
收藏资源简介:
Cardiac-CLIP是一个专为3D心脏CT图像设计的多模态基础模型,旨在解决心脏疾病诊断中的复杂问题。该模型通过大规模真实临床数据和公开数据集进行预训练,包括11,106个真实世界心脏CT扫描和5,535个胸部CT扫描,每个扫描都与放射科医生撰写的诊断报告配对。Cardiac-CLIP采用两阶段预训练策略,首先通过3D掩码自编码器进行自监督表示学习,然后通过对比学习对齐视觉和文本表示。为了全面评估Cardiac-CLIP的有效性,在多个任务上进行了广泛实验,包括心血管异常分类、信息检索和临床分析。实验结果表明,Cardiac-CLIP在多个下游任务中取得了最先进的性能,并在复杂现实世界临床分析中表现出卓越的能力,反映了其在临床决策支持方面的强大应用潜力。
Cardiac-CLIP is a multimodal foundation model specifically designed for 3D cardiac CT images, aiming to tackle complex challenges in cardiac disease diagnosis. The model is pre-trained on large-scale real-world clinical data and public datasets, including 11,106 real-world cardiac CT scans and 5,535 chest CT scans, each paired with a diagnostic report authored by radiologists. Cardiac-CLIP adopts a two-stage pre-training strategy: first performing self-supervised representation learning via a 3D masked autoencoder, then aligning visual and textual representations through contrastive learning. To comprehensively evaluate the effectiveness of Cardiac-CLIP, extensive experiments were conducted across multiple tasks, including cardiovascular abnormality classification, information retrieval, and clinical analysis. Experimental results demonstrate that Cardiac-CLIP achieves state-of-the-art performance on multiple downstream tasks and exhibits exceptional capabilities in complex real-world clinical analysis, reflecting its strong application potential in clinical decision support.




