PMILACG Model: A Predictive Model for Identifying Invasiveness of Lung Adenocarcinoma Based on High-Resolution CT-Determined Ground Glass Nodule Features

Bo Yan, Yifeng Jiang, Shijie Fu, Rong LiThe Tohoku Journal of Experimental Medicine 2024 · DOI

Abstract

Ground-glass nodules (GGNs) detected via high-resolution computed tomography (HRCT) represent a spectrum of lung adenocarcinoma (LUAD) subtypes, ranging from pre-invasive lesions to invasive adenocarcinoma. This study aimed to develop a systematic predictive model, the ILAG (Invasion of Lung Adenocarcinoma by GGN features) model, to differentiate invasive from non-invasive LUAD pre-operatively. Analyzing 1,599 GGNs from 1,189 patients, researchers identified six independent risk factors: diameter, mean CT attenuation, density uniformity, nodule-lung interface definition, spiculation, and vessel changes. The resulting ILAG model achieved an area under the curve (AUC) of 0.905, demonstrating high diagnostic accuracy for estimating LUAD invasiveness and offering a non-invasive tool to guide surgical planning.

Introduction

The management of lung adenocarcinoma (LUAD) presenting as ground-glass nodules (GGNs) on high-resolution computed tomography (HRCT) is a significant clinical challenge. GGNs can range from pre-invasive lesions, such as atypical adenomatous hyperplasia (AAH) and adenocarcinoma in situ (AIS), to invasive adenocarcinoma (IAC). Determining the level of invasiveness pre-operatively is crucial for surgical planning; non-invasive lesions may be treated with limited resection to preserve lung function, whereas invasive lesions typically require a more aggressive lobectomy. Currently, definitive diagnosis relies on post-operative pathology or intraoperative frozen sections, which can be time-consuming. There is a critical need for reliable pre-operative imaging-based predictive tools.

Methods

This retrospective study analyzed 1,599 GGNs from 1,189 patients who underwent surgical resection at Shanghai Chest Hospital between May 2020 and November 2021. Two independent radiologists evaluated 15 morphological features on HRCT. Pathological diagnosis served as the ground truth, categorizing nodules into non-invasive (n=526) and invasive (n=1,073) groups. Multivariate logistic regression was employed to identify independent risk factors and construct the ILAG predictive model.

Results

The analysis identified six independent radiological predictors of invasiveness:

  1. GGN Diameter (OR = 1.382)
  2. Mean CT Attenuation (OR = 1.007)
  3. Heterogeneous Uniformity of Density (OR = 2.151)
  4. Not Defined Nodule-Lung Interface (OR = 1.915)
  5. GGN with Spiculation (OR = 2.097)
  6. Vessel Changes (Type I OR = 1.678; Type II OR = 3.577)

Notably, Type II vessel changes—characterized by distorted or dilated vessels within the nodule—were the strongest individual morphological predictor of invasion.

ROC curve for GGN with vessel changes (AUC: 0.716).

The integrated ILAG model, combining these six features, significantly outperformed any individual metric, achieving an area under the curve (AUC) of 0.905 (95% CI: 0.890–0.919).

The ILAG model demonstrates superior predictive value with an AUC of 0.905.

Discussion

The ILAG model provides a robust, non-invasive method for estimating LUAD invasiveness pre-operatively. By leveraging standard HRCT features, it offers clinicians a systematic way to guide surgical decisions, potentially reducing the reliance on intraoperative frozen sections and helping to preserve healthy lung tissue in low-risk patients.

However, the study has limitations. As a single-center retrospective analysis of a surgical population, it may be subject to selection bias. Additionally, the assessment of morphological features involves some subjectivity between practitioners. Future work should focus on external validation across different institutions and the potential integration of AI-driven radiomics to further enhance predictive precision.

Conclusion

The ILAG model, integrating diameter, density, interface, spiculation, and vessel architecture, is a highly accurate tool for predicting lung adenocarcinoma invasiveness, facilitating personalized surgical management for patients with ground-glass nodules.

Summary updated

Abstract

The morphology of ground-glass nodule (GGN) under high-resolution computed tomography (HRCT) has been suggested to indicate different histological subtypes of lung adenocarcinoma (LUAD); however, existing studies only include the limited number of GGN characteristics, which lacks a systematic model for predicting invasive LUAD. This study aimed to construct a predictive model based on GGN features under HRCT for LUAD. A total of 1,189 surgical LUAD patients were enrolled, and their GGN-related features were assessed by 2 individual radiologists. The pathological diagnosis of the invasive LUAD was established by pathologic examination following surgery (including 1,073 invasive and 526 non-invasive LUAD). After adjustment by multivariate logistic regression, GGN diameter (OR = 1.382, 95% CI: 1.300-1.469), mean CT attenuation (OR = 1.007, 95% CI: 1.006-1.009), heterogeneous uniformity of density (OR = 2.151, 95% CI: 1.587-2.915), not defined nodule-lung interface (OR = 1.915, 95% CI: 1.384-2.651), GGN with spiculation (OR = 2.097, 95% CI: 1.519-2.896), type I (OR = 1.678, 95% CI: 1.216-2.371), and type II (OR = 3.577, 95% CI: 1.153-11.097) vessel changes were independent risk factors for invasive LUAD. In addition, a predictive model integrating these six independent GGN features was established (named as invasion of lung adenocarcinoma by GGN features (ILAG)), and receiver-operating characteristic curve illustrated that the ILAG model presented good predictive value for invasive LUAD (AUC: 0.905, 95% CI: 0.890-0.919). In conclusion, The ILAG predictive model, which integrates imaging features of GGN via HRCT, including diameter, mean CT attenuation, heterogeneous uniformity of density, not defined nodule-lung interface, GGN with spiculation, type I, and type II vessel changes, shows great potential for early estimation of LUAD invasiveness.
BibTeX
@article{yan2024pmilacg,
  author = {Bo Yan and Yifeng Jiang and Shijie Fu and Rong Li},
  title = {{PMILACG Model: A Predictive Model for Identifying Invasiveness of Lung Adenocarcinoma Based on High-Resolution CT-Determined Ground Glass Nodule Features}},
  journal = {The Tohoku Journal of Experimental Medicine},
  year = {2024},
  doi = {10.1620/tjem.2024.j078},
}
Figure