研究目的
To propose a technique for accurate and smooth 3D surface reconstruction of retinal vascular structures from 2D images for applications like computational fluid dynamics simulations and virtual interventions.
研究成果
The proposed method successfully reconstructs 3D surfaces of retinal vascular networks with smooth transitions at branching points, outperforming previous methods in terms of surface quality. It is suitable for applications in computational fluid dynamics and medical visualization. Future work should extend to 3D blood flow simulations and larger datasets.
研究不足
The method relies on accurate segmentation and correspondence matching, which may be affected by image quality. The self-calibration assumes identical focal length for two views, which might not hold in all cases. The approach is tested on a small sample size (two subjects), limiting generalizability. Computational complexity may be high for large datasets.
1：Experimental Design and Method Selection:
The methodology involves three main steps: segmentation of blood vessels using 2D matched filters, 3D skeleton reconstruction using epipolar geometry and self-calibration with Kruppa equations, and 3D surface modelling using a curvature-dependent subdivision approach. Theoretical models include the pinhole camera model, fundamental matrix estimation, and Catmull-Clark subdivision surfaces.
2：Sample Selection and Data Sources:
Retinal images were acquired from two healthy subjects (a 25-year-old male and a 20-year-old female) using a Topcon TRC-50EX fundus camera. Images were digitized via a CCD camera and transferred to a PC for analysis.
3：List of Experimental Equipment and Materials:
Topcon TRC-50EX fundus camera, CCD camera, PC for analysis. Software and algorithms for image processing, segmentation, and 3D reconstruction were implemented.
4：Experimental Procedures and Operational Workflow:
- Acquire stereo retinal images. - Segment blood vessels using 2D matched filters on the green plane of RGB images. - Extract skeletons and corresponding points. - Estimate fundamental matrix using the gold standard algorithm. - Perform self-calibration to find intrinsic parameters via simplified Kruppa equations. - Recover essential matrix and projection matrices. - Reconstruct 3D skeletons through linear triangulation. - Calculate vessel widths from segmented images. - Generate surface meshes using curvature-dependent subdivision. - Validate results through quantitative comparison of angles and visual inspection.
5：Data Analysis Methods:
Data analysis involved geometric error minimization for fundamental matrix estimation, linear equations for self-calibration, cubic spline interpolation for point matching, and curvature estimation for adaptive mesh subdivision. Statistical methods included percent error calculation for angle comparisons.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
