Predicting visceral pleural invasion in invasive adenocarcinoma with a maximum diameter ≤ 3 cm based on ¹⁸F-FDG...

To investigate the feasibility of ¹⁸F-fluorodeoxyglucose positron emission tomography/computed tomography (¹⁸F-FDG PET/CT) radiomics in preoperative prediction of visceral pleural invasion (VPI) status in invasive adenocarcinoma (IAC) with a maximum diameter ≤ 3 cm.

A total of 590 IAC patients with a maximum diameter ≤ 3 cm were enrolled and divided into training set (n = 364), validations set 1 (n = 156) and validation set 2 (n = 70). A conventional model was built based on clinical and PET/CT imaging features by logistic regression. Radiomics features extracted from CT and PET images were screened using interclass correlation coefficients, Pearson correlation analysis and the least absolute shrinkage and selection operator. These selected features were used to calculate the CT and PET rad-scores. Finally, a combined model was constructed using multivariate logistic regression.

Tumor type [odds ratio (OR): 3.258, P = 0.012], distance between tumor and pleura (OR: 0.464, P = 0.001), and maximum standardized uptake value (SUVmax) (OR: 1.109, P = 0.002) were used to construct the conventional model. Ten CT radiomics features and six PET radiomics features were used to establish the CT and PET rad-score models. The area under the curve (AUC) value of the combined model (0.824) was higher than conventional model (0.734), CT rad-score model (0.790) and PET rad-score model (0.748) in the training set, and the differences were statistically significant as tested by Delong test (P < 0.05). In the validation set 1 and validation set 2, the combined model exhibited the highest AUC values (0.835 and 0.787), and the difference between the combined model and PET rad-score model (validation set 1: 0.835 vs. 0.747, P = 0.028; validation set 2: 0.787 vs. 0.657, P = 0.043) and CT rad-score model (validation set 2: 0.787 vs. 0.694, P = 0.025) was statistically significant.

The combined model based on PET/CT radiomics is an effective and non-invasive tool for preoperative predicting VPI status in IAC patients.