一种基于生成式同异学习与原型引导的CT图像域适应肺结节分类方法、介质及设备
By employing generative similarity-dissimilarity learning and prototype-guided methods, a lung nodule classification model was constructed. This solved the problem of CT data distribution offset in cross-domain applications, achieving accurate lung nodule classification and type differentiation in different domains, and improving the generalization ability and accuracy of the classification model.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SICHUAN UNIVERSITY OF SCIENCE AND ENGINEERING
- Filing Date
- 2026-05-06
- Publication Date
- 2026-07-17
AI Technical Summary
When existing technologies are applied across domains, the distribution offset of CT data leads to a decrease in the accuracy of deep learning-based automatic classification of lung nodules. Furthermore, manually set pixel-domain geometric transformations are difficult to generate effective semantic information, affecting the accuracy and consistency of lung nodule classification.
We employ generative similarity-dissimilarity learning and prototype-guided methods. By training with generative similarity-dissimilarity learning and guiding with prototype labels, we construct a lung nodule classification model. We use the generative model to interpolate in the feature space to generate semantically invariant similar samples, and learn the classification ability of the target domain through self-supervised learning.
It achieves accurate classification of pulmonary nodules in different domains, reduces annotation costs, improves cross-domain generalization ability, can accurately distinguish nodule types, and has strong clinical application value.
Smart Images
Figure CN122156825B_ABST