基于CT图像的纹理分析鉴别肝脏实性局灶性病变

Differentiation of Solid Focal Liver Lesions:A CT-based Texture Analysis

目的 CT是鉴别肝脏实性局灶性病灶的常用检查方法,但其对不典型病灶的鉴别诊断仍有较大的经验依赖性,而纹理分析可以提供客观、定量的图像描述特征。本研究旨在探讨基于CT图像的纹理分析在肝脏实性局灶性病变鉴别诊断中的价值。资料与方法回顾性分析258例经病理证实或临床确诊的肝脏局灶性病变患者的CT图像,其中肝脏局灶性结节增生（FNH）34例,血管瘤（HEM）60例,肝细胞肝癌（HCC）60例,肝内胆管细胞癌（ICC）44例,转移瘤（MET）60例。所有患者均行腹部CT平扫与三期增强扫描。以Ma Zda软件生成CT图像的纹理特征并进行特征筛选,进行各组病灶的判别。结果 258例患者中,基于增强CT图像的纹理分析对于肝脏实性局灶性病变的鉴别诊断错判率（4.26%~37.80%）低于基于平扫图像的纹理分析（9.57%~39.02%）。对于良恶性病变的鉴别,门静脉期图像纹理分析错判率最低（13.57%）;对于FNH与HEM的鉴别,动脉期及门静脉期图像纹理分析效果相当（错判率为4.26%）;对于恶性肿瘤间的鉴别纹理分析错判率相对较高,若于恶性肿瘤间两两鉴别则错判率可降低（错判率最低为HCC与MET,约11.67%）。结论基于CT图像的纹理分析可以作为肝脏实性局灶性病灶鉴别诊断的辅助手段,尤其是FNH与HEM、良性病灶与恶性病灶、恶性病灶间的两两鉴别;其中基于三期增强扫描的纹理分析较基于平扫图像者效果更优。

Purpose CT is a common tool for the differentiation of solid focal liver lesions. However, the traditional CT assessment relies heavily on radiologists＇ experience, which leaves the differential diagnosis unfaithful. Conversely, texture analysis（TA） provides an objective and quantitative description of images. Under such circumstances, this study aims to discuss the value of TA based on both non-enhanced and triphasic contrast-enhanced CT in the differentiation of solid focal liver lesions. Materials and Methods The CT images of 258 patients with pathologically proven focal nodular hyperplasia（FNH, n=34）, hemangioma（HEM, n=60）, hepatocellular carcinoma（HCC, n=60）, intrahepatic cholangiocarcinoma（ICC, n=44）, metastasis（MET, n=60） were retrospectively analyzed. All the patients underwent non-enhanced CT and triphasic contrast-enhanced CT（CECT） scan with a standard protocol. A list of texture features was generated with free software Ma Zda for lesions＇ classification. Results The TA based on CECT had lower misclassification rate（MCR） of differentiation（4.26%-37.80%） compared with that on NECT（9.57%-39.02%） in general. In the differentiation between benign and malignant lesions, the lowest MCR was obtained on portal venous phase（13.57%）; in the differentiation between FNH and HEM, similar MCR was observed on both arterial and portal venous phases（4.26%）; in the differentiation of malignant lesions, although TA yielded a comparably higher MCR, better results were observed in the differentiation of malignant masses in pairs（the lowest MCR was 11.67% for HCC versus MET）. Conclusion The CT-based TA could serve as a supplementary tool in the differentiation of solid focal liver lesions, and it has better performance in the differential analysis of FNH and HEM, benign and malignant lesions, and malignant lesions in pairs. The triphasic CECT contains more relevant discriminatory textural information in compared with the non-enhanced CT.