肝超声造影定量分析中PCA呼吸运动模型的图像序列门控法研究

The early diagnosis of liver cancer using Contrast-Enhanced ultrasound (CEUS) imaging technique mainly depends on the observation, explaination and judgement of the enhancement level about 5 min and the corresponding change from the liver arterial phase, portal phase and delayed phase by ultrasonagraphists. It is difficult to accurately judge the little difference of enhancement change between liver malignant tumor and liver parenchyma by naked eyes. Ultrasonagraphists begin to use CEUS quantitative software to analyze the enhancement change regularity of different regions of interest during different phases. However, the clinical application of quantitative analysis is hindered greatly by respiratory motion. For reducing the effect of respiratory motion in the contrast image sequences of long phases, this research will explore the effective respiratory motion correction approach. First, the quantitative selection method will be explored to get the optimum ultrasound 2-D image and make it as the reference image for observing CEUS. Second, principal components analysis (PCA) respiratory motion model will be built based on CEUS image sequences with good respiratory regularity. The images of similar respiratory phases will be gotten from the synthetical respiratory motion curves. Then, the registration results of combing the character of contrast and tissue images and the registration results based on the character of tissue/contrast images will be compared, which will result in getting the optimum image registration mode. Finally, the accuracy and effectiveness of correction method will be demonstrated and evaluated on the clinical data. This research will help doctors improve the diagnosis efficiency of liver tumor by improving the accuracy of the liver CEUS quantitative analysis from long phase.

超声造影（CEUS）成像技术对肝癌早期诊断主要依靠医生对肝动脉相、门静脉相和延迟相持续约5分钟的增强水平及其变化的观察、解释和判断。恶性肿瘤和肝实质增强变化较小差异很难凭肉眼准确评判，超声医生常需借助CEUS定量软件分析不同感兴趣区在不同时相的增强水平变化规律。然而，呼吸运动严重阻碍长时相定量分析的临床应用。为了减少长时相造影图像序列中的呼吸运动影响，本研究对有效呼吸运动校正方法展开研究。首先探索量化选择法获取最佳的超声参考面图像；其次，针对较好呼吸运动规律的肝CEUS图像序列建立主成分分析呼吸运动模型，从合成的呼吸运动曲线获取相近呼吸相位的图像；然后，比较联合运用造影和组织图像特点配准与基于组织/造影图像特点配准的结果，获取最佳图像配准模式；最后，在临床病例数据中验证提出的校正方法准确性并评价其有效性。本研究成果通过提高肝CEUS长时相定量分析的准确性帮助医生提高肝肿瘤的诊断率。

超声造影（CEUS）成像技术对肝癌早期诊断主要依靠医生对肝动脉相、门静脉相和延迟相持续约5分钟的增强水平及其变化的观察、解释和判断。恶性肿瘤和肝实质增强变化较小差异很难凭肉眼准确评判，超声医生常需借助CEUS定量软件分析不同感兴趣区在不同时相的增强水平变化规律。然而，呼吸运动严重阻碍长时相定量分析的临床应用。为了减少长时相造影图像序列中的呼吸运动影响，本研究对有效呼吸运动校正方法展开研究。本研究先采用拉普拉斯特征映射（LE）法将原始高维超声数据降至二维空间，再采用K-means法进行聚类分析获取理想的超声参照图。随后，在18例兔肝VX2肿瘤的CEUS图像序列建立主成分分析(PCA)呼吸运动模型，尝试利用数据比例较大的主成分合成出呼吸运动曲线，然后挑选与参照图像相近相位的图像。结果：校正后图像序列的平均结构相似度和平均互相关系数均值均明显增加（P<0.001），分别为0.57±0.11和0.78±0.11。校正后图像序列的偏差值（DV）均值下降到29.9±7.02，约为原值的1/3。采用阈值设定还可进一步提高挑选的图像序列质量。该校正方法在兔的肝CEUS病例中初步验证其有效性，且操作简单，有助于提高肝肿瘤的良恶性鉴别诊断率。