Fananapazir Ghaneh, Bashir Mustafa R, Marin Daniele, Boll Daniel T
Department of Radiology, Duke University Medical Center, DUMC 3808, Durham, NC, 27710, USA.
Abdom Imaging. 2015 Jun;40(5):1203-12. doi: 10.1007/s00261-014-0276-9.
To evaluate the performance of a prototype, fully-automated post-processing solution for whole-liver and lobar segmentation based on MDCT datasets.
A polymer liver phantom was used to assess accuracy of post-processing applications comparing phantom volumes determined via Archimedes' principle with MDCT segmented datasets. For the IRB-approved, HIPAA-compliant study, 25 patients were enrolled. Volumetry performance compared the manual approach with the automated prototype, assessing intraobserver variability, and interclass correlation for whole-organ and lobar segmentation using ANOVA comparison. Fidelity of segmentation was evaluated qualitatively.
Phantom volume was 1581.0 ± 44.7 mL, manually segmented datasets estimated 1628.0 ± 47.8 mL, representing a mean overestimation of 3.0%, automatically segmented datasets estimated 1601.9 ± 0 mL, representing a mean overestimation of 1.3%. Whole-liver and segmental volumetry demonstrated no significant intraobserver variability for neither manual nor automated measurements. For whole-liver volumetry, automated measurement repetitions resulted in identical values; reproducible whole-organ volumetry was also achieved with manual segmentation, p(ANOVA) 0.98. For lobar volumetry, automated segmentation improved reproducibility over manual approach, without significant measurement differences for either methodology, p(ANOVA) 0.95-0.99. Whole-organ and lobar segmentation results from manual and automated segmentation showed no significant differences, p(ANOVA) 0.96-1.00. Assessment of segmentation fidelity found that segments I-IV/VI showed greater segmentation inaccuracies compared to the remaining right hepatic lobe segments.
Automated whole-liver segmentation showed non-inferiority of fully-automated whole-liver segmentation compared to manual approaches with improved reproducibility and post-processing duration; automated dual-seed lobar segmentation showed slight tendencies for underestimating the right hepatic lobe volume and greater variability in edge detection for the left hepatic lobe compared to manual segmentation.
评估基于MDCT数据集的全肝和肝叶分割全自动后处理原型解决方案的性能。
使用聚合物肝脏模型，通过阿基米德原理确定的模型体积与MDCT分割数据集比较，评估后处理应用的准确性。对于经机构审查委员会批准、符合健康保险流通与责任法案的研究，纳入了25名患者。体积测量性能将手动方法与自动原型进行比较，使用方差分析比较评估全器官和肝叶分割的观察者内变异性和组间相关性。定性评估分割的逼真度。
模型体积为1581.0±44.7 mL，手动分割数据集估计为1628.0±47.8 mL，平均高估3.0%，自动分割数据集估计为1601.9±0 mL，平均高估1.3%。全肝和节段体积测量显示，手动和自动测量的观察者内变异性均无显著差异。对于全肝体积测量，自动测量重复结果相同；手动分割也实现了可重复的全器官体积测量，p（方差分析）0.98。对于肝叶体积测量，自动分割比手动方法提高了可重复性，两种方法的测量差异均无显著性，p（方差分析）0.95 - 0.99。手动和自动分割的全器官和肝叶分割结果无显著差异，p（方差分析）0.96 - 1.00。分割逼真度评估发现，与其余右肝叶节段相比，I - IV/VI节段的分割不准确程度更高。
自动全肝分割显示，与手动方法相比，全自动全肝分割具有非劣效性，可提高可重复性和后处理持续时间；自动双种子肝叶分割显示，与手动分割相比，有低估右肝叶体积的轻微趋势，左肝叶边缘检测的变异性更大。
