研究目的
To develop a nonrigid 3D breast surface reconstruction pipeline using a low-cost RGBD camera to handle postural sway and improve geometric accuracy for surgical planning and aesthetic evaluation in breast cancer treatment.
研究成果
The proposed nonrigid reconstruction method effectively mitigates postural sway, producing higher geometric accuracy and better texture mapping than rigid methods. It achieves lower landmark errors and more accurate breast volume estimates, with volumes within 20 ml of gold standard in most cases, making it suitable for clinical use in breast cancer treatment planning and evaluation.
研究不足
The method has high computational time (1-2 hours per dataset), requires manual landmark picking, and may be affected by postural differences between acquisitions. Breast segmentation is challenging and error-prone, and the sti?ness parameter needs tuning to avoid overfitting. Limited to patients with distinct skin features and may not handle overly ptotic breasts well.
1：Experimental Design and Method Selection:
The method uses a two-phase globally consistent alignment with nonrigid ICP for registration, incorporating soft mobility constraints and shortest distance correspondences. It involves preprocessing (calibration, foreground masking, rigid pose estimation), graph construction (downsampling, triangulation), nonrigid alignment (bundle and global fits), spatial propagation, and model generation (MLS, Poisson reconstruction).
2：Sample Selection and Data Sources:
Data from 6 clinical patients diagnosed with early breast cancer, captured using a Microsoft Kinect sensor and a gold standard 3dMD system.
3：List of Experimental Equipment and Materials:
Microsoft Kinect sensor (first generation), 3dMD system, standard PC with Intel i7 CPU, Nvidia GeForce GTX 1050 GPU, neutral blue background, diffused studio lights.
4：Experimental Procedures and Operational Workflow:
Patients performed a 180° self-rotation while RGBD data was captured. Preprocessing included camera calibration, RGBD registration, foreground erosion. Rigid pose estimation was done with SLAM, followed by nonrigid ICP alignment in bundles and globally, then deformation propagation and fusion into a 3D model.
5：Data Analysis Methods:
Landmark error calculation using covariance matrices, breast volume estimation via polyhedral mass integrals, statistical analysis including linear regression and Bland-Altman plots.
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
