Performance of deep learning model and radiomics model for preoperative prediction of spread through air spaces in the surgically resected lung adenocarcinoma: a two-center comparative study
Open Access
- 1 December 2024
- journal article
- Published by AME Publishing Company in Translational Lung Cancer Research
- Vol. 13 (12) , 3486-3499
Abstract
Background: Spread through air spaces (STAS) in lung adenocarcinoma (LUAD) is a distinct pattern of intrapulmonary metastasis where tumor cells disseminate within the pulmonary parenchyma beyond the primary tumor margins. This phenomenon was officially included in the World Health Organization (WHO)’s classification of lung tumors in 2015. STAS is characterized by the spread of tumor cells in three forms: single cells, micropapillary clusters, and solid nests. Clinical studies have linked STAS to a poorer prognosis, higher recurrence risk, and more advanced clinicopathological staging in LUAD patients. In this study, we constructed radiomics models and deep learning models based on computed tomography (CT) for predicting preoperative STAS status in LUAD. Methods: A total of 395 (57.19±11.40 years old) patients with pathologically confirmed LUAD from two centers were enrolled in this retrospective study, in which STAS was detected in 146 patients (36.96%). The general clinical data, preoperative CT images, and the results of pathology reports of all patients were collected. Two experienced radiologists independently segmented the lesions by medical imaging interaction toolkit (MITK) software. The CT-based models only, the clinical-based models only, and the fusion model based on the two were constructed using radiomics and deep learning methods, respectively. The diagnostic performance of the different models was evaluated by comparing the area under the curve (AUC) of the subjects’ receiver operating characteristics (ROCs). Results: The deep learning model based on CT images achieved satisfactory discriminative performance in predicting STAS and outperformed the radiomics model and the clinical-radiomics model. The AUC of deep learning model was 0.918 for the internal test set and 0.766 for the external test set. The radiomics model had an AUC of 0.851 for the internal test set and an AUC of 0.699 for the external test set. The clinical-radiomics deep learning model was slightly less effective than the deep learning model (internal AUC =0.915, external AUC =0.773). Conclusions: The constructed deep learning model based on preoperative chest CT can be used to determine the STAS status of LUAD patients with good diagnostic performance and is superior to radiomics models.Keywords
This publication has 28 references indexed in Scilit:
- Radiomics signature: A potential and incremental predictor for EGFR mutation status in NSCLC patients, comparison with CT morphologyLung Cancer, 2019
- Can peritumoral radiomics increase the efficiency of the prediction for lymph node metastasis in clinical stage T1 lung adenocarcinoma on CT?European Radiology, 2019
- CT Manifestations of Tumor Spread Through Airspaces in Pulmonary Adenocarcinomas Presenting as Subsolid NodulesJournal of Thoracic Imaging, 2018
- Tumor Spread Through Air Spaces Identifies a Distinct Subgroup With Poor Prognosis in Surgically Resected Lung Pleomorphic CarcinomaChest, 2018
- Genetic and clinicopathologic characteristics of lung adenocarcinoma with tumor spread through air spacesLung Cancer, 2018
- Spread Through Air Spaces Is a Prognostic Factor in Sublobar Resection of Non-Small Cell Lung CancerThe Annals of Thoracic Surgery, 2018
- Significance of Spread Through Air Spaces in Resected Pathological Stage I Lung AdenocarcinomaThe Annals of Thoracic Surgery, 2018
- The 2015 World Health Organization Classification of Lung TumorsJournal of Thoracic Oncology, 2015
- Prognostic Impact of Intra-alveolar Tumor Spread in Pulmonary AdenocarcinomaAmerican Journal of Surgical Pathology, 2015
- Comparing the Areas under Two or More Correlated Receiver Operating Characteristic Curves: A Nonparametric ApproachBiometrics, 1988