Heatmap-Based Active Shape Model for Landmark Detection in Lumbar X-ray Images.
1Digital Health Research Division, Korea Institute of Oriental Medicine, 1672 Yuseong-daero, Yuseong-gu, Daejeon, 34054, Republic of Korea.
Journal of Imaging Informatics in Medicine
|August 5, 2024
Summary
This study enhances lumbar spine landmark detection in X-rays using a deep learning model combined with statistical shape analysis. The improved method reduces detection errors, increasing the reliability of automated diagnostic systems.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Biomedical Engineering
Background:
- Automated lumbar spine disease diagnosis relies on accurate landmark detection in X-ray images.
- Deep learning models are used for this analysis but face challenges with image noise, ambiguity, and anatomical variations, leading to detection errors.
Purpose of the Study:
- To develop a robust method for improving landmark detection accuracy in lumbar X-ray images.
- To enhance the reliability of automated systems for lumbar spine disease diagnosis.
Main Methods:
- A two-step deep learning model (Pose-Net and M-Net) was employed to detect landmarks and generate heatmap responses.
- Landmark positions were refined using heatmap responses and an active shape model incorporating statistical landmark distribution information.
Main Results:
- The proposed method significantly reduced landmark detection errors in 3600 lumbar X-ray images.
- The average maximum error decreased by 5.58% by integrating deep learning with statistical shape constraints.
- Further integration with techniques like CoordConv layers and non-directional part affinity field enhanced performance.
Conclusions:
- The combined deep learning and statistical shape model approach improves the robustness of landmark detection in lumbar X-rays.
- This advancement enhances the reliability of automated diagnostic systems, benefiting patients and medical professionals through reduced costs and improved efficiency.
