|
Objective: To investigate the Bayesian analysis of the value of CT diagnosis of solitary pulmonary nodule (SPN), and expect to provide help for the management of SPN. Methods: A retrospective collect continuous SPN cases, 352 cases (135 cases of malignant, benign 217 cases) as the training set. The use of Bayesian analysis start training set performance obtained the malignant SPN pretest and clinical and CT likelihood ratio, to calculate the each SPN vicious probability, the probability of ≥ 50% of the sentence as malignant, <50% of sentenced as benign. And to prospectively examine the Bayesian analysis of 132 cases the SPN set of test samples (61 cases of malignant and 71 benign cases) the accuracy of diagnostic performance and the predicted probability, with two high qualification and two low seniority radiologist regular reading the performance of the sheet for comparison. Results: (1) malignant SPN pretest than 0.61; (2) based on clinical and CT likelihood ratio high and low draw, better prompted the malignant SPN characterized vacuole sign, short glitches, deep leaf , better able to prompt the characteristics of benign SPN for benign calcifications mode, strengthen value <20HU, \, 93.1%, 91.5%. Test set, Bayes analysis of sensitivity, specificity, the coincidence rate, positive predictive value and negative predictive value were 88.5%, 85.9%, 87.1%, 84.4%, 89.7%, the diagnosis rate and high qualification A doctor ( 80.3% X to 2 = 2.37, P = 0.122) and acetic doctors (79.5% x to 2 = 3.12, P = 0.076) no significant difference, but higher than the low years of C doctors (74.2%, x ~ 2 = 7.05, P = 0.012) and Dr. Ding (74.2%, x 2 = 6.56, P = 0.009); (4) for non-metastatic tumor diagnosis of SPN, Bayes analysis of the area under ROC curve (Az) 0.957, greater than to Dr. Gao Nianzi group (AZ = 0.886, P = 0.003) and junior doctors (Az = 0.845, P = 0.000); (5) Bayesian analysis, high qualification A, B doctors, and the less experienced C Dr. Ding B values ??were 0.099,0.140,0.137,0.154,0.179; (6) except for the wrongful convictions of 11 cases of isolated metastases, Bayesian analysis to estimate the probability of <20% false negative rate was 1.0% (5 / 484). Conclusion: (1) associated with the SPN clinical and CT likelihood ratio can be used to guide the day-to-day read the piece; (2) Bayesian analysis is an effective diagnostic tool, can improve the performance of doctors to identify ability of SPN benign and malignant nature, especially for junior doctors help would be great; (3) Bayesian analysis forecast malignant SPN probability of high accuracy, is expected to provide a quantitative benchmark for SPN clinical decision-making.
|