[Clinical value of a differentiation prediction model for invasive lung adenocarcinoma]

Objective: To investigate the value of predicting the degree of differentiation of pulmonary invasive adenocarcinoma (IAC) based on CT image radiomics model and the expression difference of immunohistochemical factors between different degrees of differentiation of lesions. Methods: The clinicopathological data of patients with pulmonary IAC confirmed by surgical pathology in the Affiliated Huai'an First People's Hospital to Nanjing Medical University from December 2017 to September 2018 were collected. High-throughput feature acquisition was performed for all outlined regions of interest, and prediction models were constructed after dimensionality reduction by the minimum absolute shrinkage operator. Receiver operating characteristic curve was used to assess the predictive efficacy of clinical characteristic model, radiomics model and individualized prediction model combined with both to identify the degree of pulmonary IAC differentiation, and immunohistochemical expressions of Ki-67, NapsinA and TTF-1 were compared between groups with different degrees of IAC differentiation using rank sum test. Results: A total of 396 high-throughput features were extracted from all IAC lesions, and 10 features with high generalization ability and correlation with the degree of IAC differentiation were screened. The mean radiomics score of poorly differentiated IAC in the training group (1.206) was higher than that of patients with high and medium differentiation (0.969, P=0.001), and the mean radiomics score of poorly differentiated IAC in the test group (1.545) was higher than that of patients with high and medium differentiation (-0.815, P<0.001). The differences in gender (P<0.001), pleural stretch sign (P=0.005), and burr sign (P=0.033) were statistically significant between patients in the well and poorly differentiated IAC groups. Multifactorial logistic regression analysis showed that gender and pleural stretch sign were related to the degree of IAC differentiation (P<0.05). The clinical feature model consisted of age, gender, pleural stretch sign, burr sign, tumor vessel sign, and vacuolar sign, and the individualized prediction model consisted of gender, pleural stretch sign, and radiomic score, and was represented by a nomogram. The Akaike information standard values of the radiomics model, clinical feature model and individualized prediction model were 54.756, 82.214 and 53.282, respectively. The individualized prediction model was most effective in identifying the degree of differentiation of pulmonary IAC, and the area under the curves (AUC) of the individualized prediction model in the training group and the test group were 0.92 (95% CI: 0.86-0.99) and 0.88 (95% CI: 0.74-1.00, respectively). The AUCs of the radiomics group model for predicting the degree of differentiation of pulmonary IAC in the training group and the test group were 0.91 (95% CI: 0.83-0.98) and 0.87 (95% CI: 0.72-1.00), respectively. The AUCs of the clinical characteristics model for predicting the degree of differentiation of pulmonary IACs in the training and test groups were 0.75 (95% CI: 0.63-0.86) and 0.76 (95% CI: 0.59-0.94), respectively. The expression level of Ki-67 in poorly differentiated IAC was higher than that in well-differentiated IAC (P<0.001). The expression levels of NapsinA, TTF-1 in poorly differentiated IAC were higher than those in well-differentiated IAC (P<0.05). Conclusions: Individualized prediction model consisted of gender, pleural stretch sign and radiomics score can discriminate the differentiation degree of IAC with the best performance in comparison with clinical feature model and radiomics model. Ki-67, NapsinA and TTF-1 express differently in different degrees of differentiation of IAC.