Assessment of intratumor heterogeneity for preoperatively predicting the invasiveness of pulmonary adenocarcinomas manifesting as pure ground-glass nodules.

Background: The increased use of low-dose computed tomography (CT) for lung cancer screening has improved the detection of ground-glass nodules. However, as the clinical utility of CT findings to predict the invasiveness of pure ground-glass nodules (pGGNs) is currently limited, differentiating pGGNs that indicate invasive adenocarcinoma (IAC) from those that represent other histological entities is challenging. We aimed to quantify intratumor heterogeneity of lung adenocarcinomas characterized by pGGNs on CT to assess its efficacy in predicting IACs before surgery. Methods: Overall, 575 patients with persistent pGGNs and a postoperative pathological diagnosis of lung adenocarcinoma were included. To quantitatively measure intratumor heterogeneity, an intratumor heterogeneity score that incorporated local radiomics features and global pixel distribution patterns was developed. Accuracy of the preoperative prediction of pathological invasiveness was evaluated using the area under the receiver operating characteristic (ROC) curve. The performance of the intratumor heterogeneity score was compared with that of radiomics features and conventional imaging findings. Results: Conventional imaging findings yielded area under the curve values of 0.832 and 0.842 for the training and validation cohorts, respectively. The performance of imaging findings was inferior to that of radiomics, which yielded area under the curve values of 0.868 and 0.879 for the training (P=0.008) and validation (P=0.007) cohorts, respectively. Similarly, the performance of imaging findings was inferior to that of the intratumor heterogeneity score, with area under the curve values of 0.860 and 0.867 for the training (P=0.019) and validation (P=0.045) cohorts, respectively. The diagnostic performance of the intratumor heterogeneity score was comparable to that of radiomics features, with no significant difference between their ROC curves (training: P=0.635; validation: P=0.686). Conclusions: The performance of the intratumor heterogeneity score was comparable to that of radiomics features and superior to that of conventional imaging findings for the preoperative prediction of the IACs that present as pGGNs.