CT灰度直方图特征在布鲁氏菌脊柱炎和化脓性脊柱炎鉴别诊断中的价值

Value of CT grayscale histogram features in the differential diagnosis of brucella spondylitis and pyogenic spondylitis

中文关键词:  布鲁氏菌脊柱炎  化脓性脊柱炎  CT  影像组学  直方图分析

英文关键词:Brucella spondylitis  Pyogenic spondylitis  Computed tomography  Radiomics  Histogram analysis

中文摘要:

  【摘要】 目的：评估矢状位CT灰度直方图特征在布鲁氏菌脊柱炎（brucella spondylitis，BS）与化脓性脊柱炎 （pyogenic spondylitis，PS）鉴别诊断中的价值。方法：收集2018年1月～2022年12月在我院行脊柱CT检查并经病理和/或病原学证实的40例BS患者[男25例，女15例；年龄51.6±13.0岁，体重指数（body mass index，BMI）23（20，28）kg/m2），BS组]和33例PS患者[男13例，女20例；年龄50.8±16.7岁，BMI 23（20，26）kg/m2），PS组]的资料。分别在两组患者矢状位CT图像上的每一层面用3D Slicer软件平台勾画感兴趣区（region of interests，ROI）并进行灰度全域直方图分析。采用卡方检验、独立样本t检验及Mann-Whitney U检验等比较两组患者的临床资料；依次采用单因素分析、相关性分析及多因素分析等，找出两种病灶之间有显著性差异的直方图特征（包括10％百分位值、1％百分位值、25％百分位值、5％百分位值、中位数、最小值、偏度和方差等）；使用Logistic回归联合筛选的特征建模，并绘制受试者工作特征（receiver operating characteristic，ROC）曲线，计算曲线下面积（area under the curve，AUC），比较各直方图特征的鉴别能力。结果：两组患者的年龄、性别和BMI均无统计学差异（P>0.05）。两组CT全域灰度直方图分析参数中，10％百分位值、1％百分位值、25％百分位值、5％百分位值、中位数、最小值、偏度和方差等8个特征的差异均有统计学意义（P<0.05），10％百分位值的诊断效能最佳，其AUC值为0.824、特异度为0.893。联合模型AUC值为0.860、特异度为0.946。结论：基于CT灰度直方图10%百分位值及联合模型能有效鉴别PS和BS，可为临床鉴别两种疾病提供依据。

英文摘要:

  【Abstract】 Objectives: To evaluate the values of sagittal CT image histogram features in the differential diagnosis of brucella spondylitis(BS) and pyogenic spondylitis(PS). Methods: The data of 40 BS patients[25 males, 15 females; age: 51.6±13.0 years old; body mass index(BMI): 23(20, 28)kg/m2, the BS group] and 33 PS patients[13 males, 20 females; age: 50.8±16.7 years old; BMI: 23(20, 26)kg/m2, the PS group] who underwent CT examination of the spine in our hospital and were confirmed through pathology and/or etiology were collected. The region of interest(ROI) was delineated on each level of the sagittal CT images of the two groups of patients by using the 3D Slicer platform and grayscale global histogram analysis was performed. The clinical data were compared using chi square test, independent sample t-test, and Mann Whitney U test between the two groups of patients; Univariate analysis, correlation analysis, and multivariate analysis were used in sequence to identify the histogram features with significant differences between the two groups(including 10% percentile, 1% percentile, 25% percentile, 5% percentile, median, minimum, skewness, and variance); Logistic regression and the screened features were combined for modeling, and receiver operating characteristic(ROC) curves were drawn and areas under the curve(AUC) were calculated to compare the discriminative ability of each histogram feature. Results: There was no statistically significant difference in age, gender, and BMI between the two groups of patients(P>0.05). Among the histogram parameters, 10% percentile value, 1% percentile value, 25% percentile value, 5% percentile value, median, minimum value, skewness, and variance were statistically different between the two groups(P<0.05). The 10% percentile value displayed the best diagnostic performance, with an AUC value of 0.824 and a specificity of 0.893. The combined model had an AUC value of 0.860 and a specificity of 0.946. Conclusions: Based on 10% percentile value of CT grayscale histogram and joint model, PS and BS can be distinguished effectively, providing a basis for accurately distinguishing the two diseases in clinical practice.