摘要
目的基于氟代脱氧葡萄糖(18F-FDG)正电子发射断层扫描成像/计算机断层扫描成像(PET/CT)图像纹理参数构建预测模型并诊断不同病理类型肺癌的临床价值。方法选择2020年6月至2021年6月于我院收治的肺部疾病患者199例,根据病理结果分为良性组及恶性组,恶性组根据病理类型分为肺腺癌组、肺鳞癌组、其他组。所有患者均行18F-FDG PET/CT检查,选择肺部病灶感兴趣区(ROI)、采集频度、峰度、对比性、粗糙度等13项18F-FDG PET/CT图像纹理参数,比较各组患者纹理参数水平,采用二分类Logisitic回归分析影响肺癌的纹理参数。采用受试者工作特征曲线(ROC)分析单一、多影响组学模型的曲线下面积(AUC)、灵敏度、特异度。结果病理检查确诊良性肺部疾病101例,肺炎占48.5%;确诊肺癌98例,肺鳞癌占59.2%。本实验共提取13项纹理参数,恶性组角二阶矩(ASM)、频度、峰度、对比性、粗糙度明显低于良性组,均值、标准差、偏度、熵、相关性、对比度、复杂度、强度明显高于良性组(P<0.05)。二分类Logistic回归分析发现,ASM、峰度、均值、粗糙度、复杂度为恶性肺部疾病发生影响因素(P<0.05)。3组肺癌患者ASM、峰度、均值、粗糙度、复杂度差异有统计学意义(P<0.05)。以ASM、峰度、均值、粗糙度、复杂度为指标应用支持向量机(SVM)构建影像组学模型,ROC曲线分析发现,纹理参数预测模型、医师诊断联合诊断不同病理类型肺癌AUC为0.891。结论18F-FDG PET/CT图像纹理参数在判断肺部疾病良恶性、肺癌病理类型方面具有积极作用,通过纹理参数构建预测模型诊断不同病理类型肺癌较好的预测价值。
Objective To construct a prediction model based on 18F-fluorodeoxyglucose(18F-FDG)positron emission tomography imaging/computed tomography imaging(PET/CT)image texture parameters and diagnose the clinical value of different pathological types of lung cancer.Methods One hundred and ninety-nine patients with lung diseases admitted to our hospital from June 2020 to June 2021 were selected and divided into benign and malignant groups according to pathological findings,and the malignant group was divided into lung adenocarcinoma group,lung squamous carcinoma group,and other groups according to pathological types.All patients underwent 18F-FDG PET/CT examination,the region of interest(ROI)of lung lesions was selected,the 1318F-FDG PET/CT image texture parameters such as frequency,kurtosis,contrast,and roughness,compared the levels of texture parameters among patients in each group were collected,and the texture parameters affecting lung cancer using dichotomous logisitic regression were analyzed.The receiver operator characteristic curve(ROC curve)was used to analyze the area under the curve(AUC),sensitivity,specificity and accuracy bounds of single and multi-influence histological models.Results The pathological examination of this experiment confirmed the diagnosis of benign lung disease in 101 cases,with pneumonia accounting for 48.5%,and confirmed lung cancer in 98 cases,with squamous lung cancer accounting for 59.2%.A total of 13 texture parameters were extracted in this experiment,and the ASM,frequency,kurtosis,contrast,and roughness were significantly lower in the malignant group than in the benign group,and the mean,standard deviation,skewness,entropy,correlation,contrast,complexity,and intensity were significantly higher than in the benign group(P<0.05).Dichotomous logistic regression analysis revealed that ASM,kurtosis,mean,roughness,and complexity were factors influencing the occurrence of malignant lung disease(P<0.05).ASM,kurtosis,mean value,roughness,and complexity were significantly different among the three groups of lung cancer patients(P<0.05).SVM was applied to construct the imaging histological model with ASM,kurtosis,mean value,roughness,and complexity as indicators,ROC curve analysis revealed that the AUC of the combined texture parameter prediction model and physician diagnosis for diagnosing different pathological types of lung cancer was 0.891.Conclusion 18F-FDG PET/CT image texture parameters have positive effects in determining benign and malignant lung diseases and pathological types of lung cancer,and the predictive value of constructing a predictive model for diagnosing different pathological types of lung cancer by texture parameters is better.
作者
李超
刘玲玲
曹宏
彭德智
Li Chao;Liu Lingling;Cao Hong;Peng Dezhi(Department of Nuclear Medicine,Xi′an International Medical Center Hospital,Shaanxi 710000,China;不详)
出处
《山西医药杂志》
CAS
2023年第3期188-192,共5页
Shanxi Medical Journal
