研究目的
Introducing a measure for segmentation uncertainty in the form of segmentation error margins to allow a 'fully informed' comparison between extracted boundaries of related ROIs and more meaningful statistical analysis.
研究成果
The proposed method for segmentation sampling using MCMC for the construction of segmentation uncertainty margins is a promising tool for medical imaging analysis, particularly for comparative longitudinal and cross-sectional studies and user-guided segmentation.
研究不足
The accuracy of segmentation error margins cannot be validated directly, hence ROC curves are used as a proof of concept.
1：Experimental Design and Method Selection:
The methodology involves a novel technique for segmentation sampling in the Fourier domain and Markov Chain Monte Carlo (MCMC) for constructing segmentation error margins.
2：Sample Selection and Data Sources:
The method was tested on 50 MR brain scans, with manual segmentations of four structures provided for each scan.
3：List of Experimental Equipment and Materials:
A
4：5-T General Electric Scanner was used to acquire the MR images. Experimental Procedures and Operational Workflow:
The algorithm was applied to 2D central slices of the volumes in a leave-one-out manner, where 49 image-segmentations pairs were used for training.
5：Data Analysis Methods:
ROC curves were used to evaluate the method, with precision and recall scores calculated with respect to expert annotations.
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
