WCLC 2025 - Posters & ePosters-P1.04.27 Radiologic Features Identification of Pathologic Tumor Invasion in Pure Ground-Glass Nodules

Pdf Summary

This study by Zhu et al. developed a CT-based radiologic ternary classification model to predict pathological invasiveness of non-solid lung nodules (NSNs), which are increasingly detected and present challenges of overdiagnosis and overtreatment. The model aims to differentiate preinvasive lesions, minimally invasive adenocarcinoma (MIA), and invasive adenocarcinoma (IAC).<br /><br />Using chest CT images from 1,683 patients with NSNs, various radiologic features were extracted before surgical resection. Significant predictors identified include nodule size, number of intranodular vessels, mean CT attenuation, uniformity of density, spiculation, lobulation, pleural retraction, bubble lucency, and air bronchogram. Statistical analysis demonstrated strong associations between these variables and pathological subtypes.<br /><br />Three classification models were compared: Model 1 (nodule size only), Model 2 (size plus CT attenuation), and Model 3 (full partial proportional odds model incorporating multiple features). Model 3 showed the highest diagnostic performance with an area under the curve (AUC) of 0.91 (95% CI: 0.89-0.93) and a Brier score of 0.119, outperforming simpler models. Optimal cutoff values were identified at 7.5 mm/–595 Hounsfield units and 10.5 mm/–512 HU to distinguish between the three pathological categories.<br /><br />The research confirms that integrated CT radiologic features can provide a robust non-invasive method to predict invasiveness in NSNs, potentially informing clinical decisions regarding follow-up versus resection strategies. Future directions include conducting multi-institutional prospective studies with long-term follow-up and integrating artificial intelligence to further enhance predictive accuracy.<br /><br />In summary, this work presents a validated radiological model with excellent diagnostic efficacy (C-index 0.92) for accurately stratifying NSNs by pathologic invasiveness, supporting personalized management of lung cancer risk and possibly reducing overtreatment.

Asset Subtitle

Yeqing Zhu

Meta Tag

Speaker Yeqing Zhu

Topic Screening and Early Detection

Keywords

CT-based classification

non-solid lung nodules

pathological invasiveness

radiologic features

minimally invasive adenocarcinoma

invasive adenocarcinoma

nodule size

CT attenuation

diagnostic model

lung cancer risk stratification