A method and system for bone image segmentation and measurement based on geometric perception deep learning

By employing a bone image segmentation method based on geometric perception deep learning, the problems of strong subjectivity and topological misjudgment in manual measurement are solved, achieving efficient and accurate bone quality assessment and continuous gradient capture, thereby improving the accuracy and efficiency of osteoporosis screening.

CN122415651APending Publication Date: 2026-07-17NORTHEASTERN UNIV CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
NORTHEASTERN UNIV CHINA
Filing Date
2026-06-05
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing technologies for manually measuring osteoporosis are subject to strong subjectivity, low efficiency, and cannot capture continuous longitudinal bone loss gradients. Traditional deep learning models are prone to topological misjudgments in skeletal structure segmentation.

Method used

A skeletal image segmentation method based on geometric perception deep learning is adopted, including a macroscopic localization network, a semantic segmentation network, and anatomical prior processing. By generating an accurate skeletal segmentation mask, extracting two-dimensional continuous morphological indicators, and applying intensity-weighted scoring and bidirectional topological inclusion strategies, the problems of topological violations and boundary leakage are overcome.

Benefits of technology

It achieves efficient and accurate bone quality assessment, significantly improves Dice similarity coefficient and classification accuracy, can output continuous bone loss gradient without human intervention, has high clinical diagnostic consistency, and is suitable for large-scale screening.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

本发明公开了一种基于几何感知深度学习的骨骼影像分割与测量方法及系统,属于医疗图像处理技术领域。针对现有骨质量评估依赖人工测量且标准深度学习模型易产生拓扑错误的问题,本发明方案为:首先获取X射线影像并由定位网络提取包含目标骨骼的感兴趣区域;利用双通道语义分割网络初步预测皮质骨与髓腔;应用基于解剖学先验的强度加权评分和双向拓扑包含策略进行后处理以生成精确分割掩膜;最后基于掩膜沿骨骼纵轴提取二维连续形态学指标,并自动输出一维临床分型。本发明有效克服了传统网络的拓扑越界错判,消除了人工主观偏差,适用于基于常规影像的高效、高鲁棒性大规模骨质量机会性筛查。
Need to check novelty before this filing date? Find Prior Art