CT study of radiomics features predicting Ki-67 expression level in peripheral lung cancer

Abstract Background The association between radiomics features of peripheral lung cancer and Ki-67 expression level remains unclear. To develop a radiomics signature based on enhanced CT arterial phase image to estimate the expression situation of Ki-67 in peripheral lung cancer. Methods A total of 117 peripheral lung cancer patients, who underwent contrast-enhanced CT scan in our hospital from May 2016 to November 2019, including 43 males and 74 females, aged 35 to 79 years old (median 54 years old). All the peripheral lung cancers were confirmed by histopathological and took in Ki-67 expression situation detection within 2 weeks after CT inspection, including 63 cases of Ki-67 low expression and 54 cases of Ki-67 high expression, which were retrospectively analyzed and were divided into training (n = 82) and validation cohorts (n = 35) in a ratio of 7:3. ITK-SNAP was used to manually outline the total tumor volume data of lung cancer on CT arterial phase images, and the radiomics features were extracted by A.K software. LASSO regression model was used to further screen features and construct radiomics labels, and the radiomics score of each patient was calculated, and then multi-factor logistic regression analysis was performed combined with clinical information to screen out independent risk factors for predicting Ki-67 levels. The predictive accuracy of the radiomics signature was quantified by the area under the curve (AUC) of receiver operator characteristic (ROC) curve in both the training and validation cohorts. The Hosmer-Lemeshow test was performed to evaluate the calibration degree of the radiomics. We performed decision curve analysis (DCA) to assess the clinical usefulness of the radiomics signature. Results Seven radiomics features were chosen from 396 candidate features to build a radiomics label that significantly correlated with Ki-67 expression level. The model showed good calibration and discrimination in the training cohort, with an AUC of 0.844 (95%CI: 0.725–0.964), sensitivity of 93% and specificity of 71%, calibration degree of 0.709. In the validation cohort, AUC was 0.881 (95%CI: 0.756–0.954), sensitivity was 91%, and specificity was 75%, calibration degree of 0.950. Univariate logistic regression analysis showed that there were no conspicuous differences in gender, age and smoke between the high and low Ki-67 expression (P > 0.05). Using multivariate logistic regression model, radiomics signature were considered to be independent predictor of Ki-67 expression level in peripheral lung cancer. DCA for the radiomics signature in the training cohort showed that if the threshold probability was between 0.03 and 0.63, then using the radiomics signature to predict Ki-67 expression situation added more benefit than treating either all or no patients. Conclusion The radiomics signature based on enhanced CT arterial phase image is benefitial to predict the expression of Ki-67 in peripheral lung cancer, which can assess the invasiveness and prognosis for peripheral lung cancer noninvasively.