一个基于PET/CT的孤立性肺结节恶性概率的预测模型

A PET/CT-BASED PREDICTION MODEL FOR MALIGNANCY PROBABILITY OF SOLITARY PULMONARY NODULES

为减少在肺结节良恶性的临床诊断过程中主治医生的临床经验、当前状态等主观因素对诊断结果的影响,提出一种基于PET/CT的孤立性肺结节恶性病变概率的辅助预测模型来尽可能消除诊断过程中的不确定因素,提高诊断的准确度。模型采用单因素方差分析法逐个分析肺结节的PET和CT各个征象,确定与肺结节恶性病变具有显著性相关的因素;利用多因素logistic分析法分析各个影响因子之间的相关性,并建立预测肺结节恶性病变概率的回归方程。通过利用Hosmer-Lemeshow检验和最大似然比检验对模型的拟合度进行检验,结果表明,该模型具有较高的拟合度。同时将该模型与Mayo和VA模型相比,将预测概率0.54作为良恶性诊断的临界时,该模型在预测准确度和敏感度方面表现出比较高的优势。

To reduce the effect of the subjective factors in the benign and malignancy diagnosis of solitary pulmonary nodules（ SPNs） such as clinical experience and current status of doctors,in this paper we proposed a PET / CT-based auxiliary prediction model for malignant lesion probability of SPNs to eliminate as much as possible the uncertainty factors in diagnoses process and to improve the accuracy of diagnoses.The model employs univariate analysis of variance method to analyse each symptom of PET and CT of SPNs one by one to identify the factors clearly relevant to the malignant lesion of SPN. And then the model uses multi-factor logistic analysis method to analyse the relevance between each identified factor,and develops a regression equation predicting SPN＇s malignant lesion probability. The fitness of the model has been tested by the Hosmer-Lemeshow and maximum likelihood ratio methods,results showed that the model had higher fitness. In addition,we compared the model with Mayo model and VA model,it demonstrated higher advantages in terms of the accuracy and sensitivity of prediction when the prediction probability 0. 54 was set as the critical diagnosis of benign and malignant.