CT-RATE-LT
收藏资源简介:
CT-RATE-LT数据集由中国科学技术大学的研究团队创建,旨在解决3D医学图像分析中的零样本诊断问题。该数据集包含33个异常类型,其中15个是代表性不足的异常类型,扩展了原有的CT-RATE数据集。数据集通过结合临床报告和3D CT图像,利用大语言模型进行语义总结,提取高层次语义信息,并通过跨模态知识交互模块进行特征对齐。该数据集的应用领域主要集中在医学图像分析,特别是针对罕见疾病的零样本诊断,旨在提高模型在长尾分布数据上的泛化能力。
The CT-RATE-LT dataset was constructed by a research team from the University of Science and Technology of China, with the goal of addressing the zero-shot diagnosis problem in 3D medical image analysis. This dataset encompasses 33 anomaly categories, 15 of which are underrepresented, thus expanding the original CT-RATE dataset. By integrating clinical reports and 3D CT images, the dataset utilizes Large Language Models (LLMs) to conduct semantic summarization, extract high-level semantic information, and perform feature alignment via cross-modal knowledge interaction modules. Its applications are primarily focused on medical image analysis, especially zero-shot diagnosis for rare diseases, aiming to enhance the generalization capability of models on long-tailed distributed data.

- 1Bridged Semantic Alignment for Zero-shot 3D Medical Image Diagnosis中国科学技术大学 · 2025年



