Abstract
Per the 2021 World Health Organization (WHO) Classification of Thoracic Tumors, poorly differentiated invasive nonmucinous adenocarcinoma (INMA) demonstrates aggressive clinicopathological behavior characterized by lymphatic invasion, correlates with poor prognosis, and necessitates modifications to surgical planning even in early-stage adenocarcinoma. This study aimed to investigate the predictive value of radiomics features of intratumoral and peritumoral microenvironment in poorly differentiated INMA in the lung. A total of 451 patients with INMA were collected from three hospitals. They were divided into the train cohort (173 grade 1/2; 116 grade 3), internal test cohort (89 grade 1/2; 35 grade 3) and external test cohort (26 grade 1/2; 12 grade 3). The logistic regression analysis was used to establish the clinical and radiomic models. The receiver operating characteristic (ROC) curve and the decision curve analysis (DCA) were used to assess diagnostic performance of different models. The internal test dataset (124 patients) was used to evaluate the radiologists' performance without and with the assistance of optimal model. Nodule attenuation, consolidation size and consolidation-to-tumor ratio (CTR) meaning the ratio of maximum consolidation to tumor dimensions on axial imaging were independent predictors of poorly differentiated INMA (p < 0.05). In the internal test cohort, the area under the curve (AUC) values of the clinical model, the intratumor radiomic model, the combined 3 mm radiomic model, and the combined 5 mm radiomic model were 0.875(95%CI: 0.811-0.938), 0.882(95%CI: 0.807-0.957), 0.907(95%CI: 0.844-0.969) and 0.858(95%CI: 0.783-0.933), respectively. Corresponding results in the external test cohort showed moderate performance across all models: clinical model (AUC: 0.760, 95%CI: 0.603-0.916), intratumor radiomic model (AUC: 0.760, 95%CI: 0.580-0.939), and combined 3 mm/5 mm radiomic model(AUC: 0.772, 95%CI: 0.593-0.952; AUC: 0.766; 95%CIs: 0.580-0.953). Radiologists had higher diagnostic performance and confidence score with the aid of the combined 3 mm radiomic model. The combined 3 mm radiomic model of non-enhanced CT can identify poorly differentiated INMA with excellent performance and improve radiologists to achieve better diagnostic performance.