Division of Image Processing, Department of Radiology, Leiden University Medical Center, Albinusdreef 2, Zone C2-S, PO Box 9600, 2300 RC Leiden, The Netherlands.
Acad Radiol. 2010 Nov;17(11):1375-85. doi: 10.1016/j.acra.2010.06.015.
Derivation of diagnostically relevant parameters from first-pass myocardial perfusion magnetic resonance images involves the tedious and time-consuming manual segmentation of the myocardium in a large number of images. To reduce the manual interaction and expedite the perfusion analysis, we propose an automatic registration and segmentation method for the derivation of perfusion linked parameters.
A complete automation was accomplished by first registering misaligned images using a method based on independent component analysis, and then using the registered data to automatically segment the myocardium with active appearance models. We used 18 perfusion studies (100 images per study) for validation in which the automatically obtained (AO) contours were compared with expert drawn contours on the basis of point-to-curve error, Dice index, and relative perfusion upslope in the myocardium.
Visual inspection revealed successful segmentation in 15 out of 18 studies. Comparison of the AO contours with expert drawn contours yielded 2.23 ± 0.53 mm and 0.91 ± 0.02 as point-to-curve error and Dice index, respectively. The average difference between manually and automatically obtained relative upslope parameters was found to be statistically insignificant (P = .37). Moreover, the analysis time per slice was reduced from 20 minutes (manual) to 1.5 minutes (automatic).
We proposed an automatic method that significantly reduced the time required for analysis of first-pass cardiac magnetic resonance perfusion images. The robustness and accuracy of the proposed method were demonstrated by the high spatial correspondence and statistically insignificant difference in perfusion parameters, when AO contours were compared with expert drawn contours.
从首次通过心肌灌注磁共振图像中推导出具有诊断意义的参数，需要对大量图像中的心肌进行繁琐且耗时的手动分割。为了减少手动交互并加快灌注分析速度，我们提出了一种用于推导出灌注相关参数的自动配准和分割方法。
通过基于独立成分分析的方法对未对准的图像进行首次配准，从而实现完全自动化，然后使用已注册的数据，通过主动外观模型自动分割心肌。我们使用 18 项灌注研究（每项研究 100 张图像）进行验证，基于点到曲线误差、骰子指数和心肌内相对灌注上升斜率，将自动获取的（AO）轮廓与专家绘制的轮廓进行比较。
18 项研究中有 15 项研究的分割结果可通过肉眼观察得出。AO 轮廓与专家绘制轮廓的比较结果为 2.23 ± 0.53mm 和 0.91 ± 0.02（点到曲线误差和骰子指数）。手动和自动获得的相对上升斜率参数之间的平均差异无统计学意义（P =.37）。此外，每片的分析时间从 20 分钟（手动）减少到 1.5 分钟（自动）。
我们提出了一种自动方法，显著减少了首次通过心脏磁共振灌注图像分析所需的时间。通过比较 AO 轮廓与专家绘制的轮廓，得出了具有高空间对应性且灌注参数无统计学差异的结果，证明了所提出方法的稳健性和准确性。
