一种基于超导心磁图仪的病灶定位方法及系统
By using an improved VM-UNet encoder and frequency domain enhanced feature spectrum technology, combined with an adaptive threshold segmentation algorithm, the problem of unclear lesion boundaries in superconducting magnetocardiography lesion localization was solved, achieving high-precision and low-resource-consumption lesion segmentation results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING SQUID QUANTUM TECH
- Filing Date
- 2026-02-25
- Publication Date
- 2026-07-17
AI Technical Summary
Existing superconducting magnetocardiography lesion localization methods struggle to effectively distinguish lesions from diffuse background areas when processing alone in the spatial domain. Traditional CNN models are difficult to achieve fine and robust lesion segmentation in weak edge and noisy regions. Transformer-type networks consume huge amounts of computational resources and are not suitable for clinical deployment.
An improved VM-UNet encoder is used, which combines a visual state space module and a two-dimensional selective scanning mechanism to extract global contextual features. Through multi-scale frequency domain information compensation branches and learnable spectral attention filters, the high-frequency amplitude and phase components related to lesions are enhanced. Furthermore, the features are fused with spatial backbone features through a residual connection mechanism and combined with a pixel-level adaptive threshold segmentation algorithm to generate a fine segmentation mask for lesions.
It improves the accuracy of lesion boundary reconstruction, reduces missegmentation and boundary blurring, improves the segmentation accuracy under complex magnetic field interference conditions, reduces the risk of artifacts and small noise areas being misjudged as lesions, and maintains efficient utilization of computing resources.
Smart Images
Figure CN121937534B_ABST