Development and validation of intratumoral and peritumoral CT-based radiomics nomograms for predicting malignancy and invasiveness of pulmonary ground-glass nodules: a retrospective study

Background: Currently, it is still very challenging for radiologists to visually distinguish between benign and malignant ground-glass nodules (GGNs) and predict the degree of invasiveness of GGNs directly from computed tomography (CT) images. Radiomics is a newly emerging technique that can transform digital medical images into numerous quantitative features revealing pathophysiology. The present study aimed to evaluate the predictive value of intratumoral and peritumoral CT-based radiomics and the radiomics nomogram for determining the benign or malignant status of pulmonary GGNs and the invasiveness of malignant GGNs.

Methods: A total of 1,372 GGNs from 1,280 patients with preoperative CT scans were included in this retrospective study. Among these, 212 GGNs were diagnosed with inflamed or infected benign lesions, 31 with atypical adenomatous hyperplasia (AAH), 257 with adenocarcinoma in situ (AIS), 559 with minimally invasive adenocarcinoma (MIA), and 313 with invasive adenocarcinoma (IAC). Malignant GGNs were classified according to clinical management strategies, with AAH, AIS, and MIA categorized as non-invasive lesions, and IAC as an invasive lesion. Two types of regions of interest (ROIs) were annotated from the CT images: gross tumor volume (GTV) and peritumoral volume (5 and 10 mm around the tumor, PTV₁ and PTV₂). Radiomics models, a clinical factor model, and a combined radiomics nomogram integrating the radiomics score with independent clinical predictors were constructed and compared.

Results: For malignancy predictions, the GTV + PTV₁ + PTV₂ radiomics model and the PTV₁ radiomics model achieved the highest performance in the training set and testing set, with areas under the curve (AUCs) of 0.840 and 0.796, respectively. The combined radiomics nomogram demonstrated excellent discrimination in both the training set and testing set, with AUCs of 0.870 and 0.817, respectively. Moreover, the DeLong test showed that in both the training and validation cohorts, the AUC values of the comprehensive nomogram were significantly different from those of the clinical factor model, GTV radiomics model, PTV₂ radiomics model, and GTV + PTV₁ + PTV₂ radiomics models (P<0.05). For invasiveness predictions, the GTV + PTV₁ + PTV₂ radiomics model exhibited the highest performance in the training set and testing set, with AUCs of 0.956 and 0.954, respectively. Similarly, the combined radiomics nomogram showed excellent discriminatory performance in both the training set and testing set, with AUCs of 0.962 and 0.957, respectively. Furthermore, the DeLong test demonstrated that the AUC values of the comprehensive nomogram were significantly different from those of the clinical factor model and the PTV₂ radiomics model in both cohorts (P<0.05).

Conclusions: The radiomic signatures of both GTV and PTV₁ show strong predictive capabilities for the malignancy and invasiveness of GGNs, with the PTV₁ radiomics model performing comparably to, or even surpassing, that of the GTV radiomics model. The proposed nomogram demonstrated the best performance in assessing malignancy and invasiveness in patients with GGNs. It may serve as a powerful tool to assist clinicians in formulating personalized treatment strategies.