交互式分割软件在肝脏局灶性病灶CT图像中的应用初探

Application of Interactive CT Image Segmentation Software in Focal Liver Lesions

【目的】探讨交互式分割软件在肝脏局灶性病变CT图像中的应用。【方法】建立基于机器学习的肝脏局灶性病灶的交互式分割软件,通过学习判别式模型,对每一个像素进行病灶/非病灶的判别,从而实现区域分割。软件对PhotoshopCS4建立的理想单幅及序列ROI图像模型进行分割后,进而在肝脏局灶性病灶的CT图像上进行分割。【结果】理想的单幅及序列ROI模型中,前景随机选择不同灰度值及直径的ROI区时,均可精确分割出所有ROI区。单幅及序列CT图像分割中,前景分别选择肝癌消融后肿瘤完全灭活(低密度)及肝脏局灶性结节增生(高密度)病灶,软件正确分割病灶与周围正常组织区域;以Photoshop人工分割作为金标准,比较软件单独及序列分割的面积,差异无统计学意义(P>0.05)。【结论】基于判别模型学习的交互式分割软件初步成功应用于肝脏局灶性病灶CT图像的分割。

[Objective] To investigate the application of interactive segmentation software in CT images of focal liver lesions.[Methods] This software is based on machine learning interactive segmentation.It achieves region segmentation by learning discriminant model for each pixel in the lesion or non-lesion discrimination.Segmentation efficiency was tested on the ideal single or sequential （Religion of Interest） ROI model established by Photoshop CS4,and segmentation in CT images of focal liver lesions was performed.[Results] In ideal single or sequential ROI model,this software would accurately segment all ROI area despite the gray values and diameter of the ROI.In single and/or sequential CT images,once the ablation lesion was selected,the software would accurately segment the region of the lesion and surrounding normal tissue.Compared with Photoshop manual segmentation,there is no statistically significant difference between the gold standard and the single or sequential segmentation （P 〉 0.05）.[Conclusion] This interactive segmentation software based on discriminant model learning has been preliminary successfully applied to CT images of focal liver lesions.