AUTHOR=Wang Bing , Zhang Hui , Li Wei , Fu Siyun , Li Ye , Gao Xiang , Wang Dongpo , Yang Xinjie , Xu Shaofa , Wang Jinghui , Hou Dailun 

TITLE=Neural network-based model for evaluating inert nodules and volume doubling time in T1 lung adenocarcinoma: a nested case−control study

JOURNAL=Frontiers in Oncology

VOLUME=Volume 13 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2023.1037052

DOI=10.3389/fonc.2023.1037052

ISSN=2234-943X

ABSTRACT=Objective: The purpose of this study is to establish a model for judging inert nodules and predicting nodule volume multiplication based on deep learning, to identify inert nodules according to CT images and clinical feature analysis of pulmonary nodules, to predict nodule volume doubling time, to assist in the clinical judgment of nodule properties, and to achieve scientific management of nodules. Methods: A total of 201 patients with T1 lung adenocarcinoma  were analyzed retrospectively, and pulmonary nodule information was predicted by the AI pulmonary nodule auxiliary diagnosis system. The nodules were divided into two groups: inert nodules (VDT>600 d, n=152) and non-inert nodules (VDT<600d, n=49). Then, taking the clinical and imaging features obtained at the first examination as predictive variables, the inert nodule judgment model (INM) and volume doubling time estimation model (VDTM) were constructed based on the neural network system of deep learning. The performance of the INM model was evaluated by the area under curve (AUC) obtained from the receiver operating characteristic (ROC) analysis, and the performance of the VDTM model was evaluated by R2 (determination coefficient).
Results: The accuracy of INM in internal validation was 81.13%, and in external verification, it was 77.50%. The ROC of INM for internal verification was 0.7707 (95% CI 0.6779-0.8636), and the ROC for external verification was 0.7700 (95% CI 0.5988-0.9412). INM is effective in identifying inert pulmonary nodules; the internal verification of R2VDTM is 0.8008, and the external verification is 0.6268. VDTM shows moderate performance in estimating VDT, which can provide some reference in patients' first examination and consultation.
Conclusion: The inert nodule judgment model based on deep learning and VDTM can help radiologists and clinicians distinguish between inert nodules and predict nodule growth time to accurately treat patients with pulmonary nodules.