MedNexus
2021年 · 第101卷第07期
MedNexus
- 全部
- 专家论坛
- 人工智能在医学影像领域的应用
- 临床研究
- 基础研究
- 病例报告
- 继续教育园地
- 综述
- 文献速览
Pancreatic segmentation is widely used in surgical planning, diabetes assessment and detection and analysis of pancreatic tumors. Because the shape, size and density of the pancreas vary greatly, and it is closely related to adjacent organs such as intestine and blood vessels, it is still a difficult problem to accurately segment the pancreas on CT images. In order to understand the factors affecting the deep learning pancreatic segmentation algorithm, this study evaluated the segmentation effect of 82 venous abdominal CT images from public datasets, analyzed from three aspects: population and clinical data, CT technical factors and CT imaging performance, and used Dice similarity coefficient (DSC) to evaluate the segmentation accuracy. The results showed that the mean DSC of pancreatic segmentation was 78% ± 8%. Factors significantly associated with pancreatic segmentation effects included patient body mass index (r=0.34,P<0.01), abdominal visceral fat (r=0.51,P<0.000 1), pancreatic volume (r=0.41,P=0.001), standard deviation of intrapancreatic CT values (r=0.30,P=0.01), the median and mean of CT values at the critical 5 mm of the pancreas (r=-0.53,P<0.000 1;r=-0.52,P<0.000 1), CT Image pixel values (r=0.31,P=0.004)。 There was no significant correlation between DSC and height, sex and mean CT values of pancreas.
本期目次

