Baseline whole-lung CT features deriving from deep learning and radiomics: prediction of benign and malignant pulmonary ground-glass nodules

• DOI: 10.3389/fonc.2023.1255007
• Publication Date: 2023-08-18T07:57:48Z
• Authors: Wenjun Huang; Heng Deng; Zhaobin Li; Zhanda Xiong; Taohu Zhou; Yanming Ge; Jing Zhang; Wenbin Jing; Yayuan Geng; Xiang Wang; Wenting Tu; Peng Dong; Shiyuan Liu; Li Fan
• Keywords: Data set

Objective To develop and validate the model for predicting benign malignant ground-glass nodules (GGNs) based on whole-lung baseline CT features deriving from deep learning radiomics. Methods This retrospective study included 385 GGNs 3 hospitals, confirmed by pathology. We used 239 Hospital 1 as training internal validation set; 115 31 2 external test sets 2, respectively. An additional 32 stable with more than five years of follow-up were set 3. evaluated clinical morphological at chest extracted radiomics simultaneously. Besides, image are further assisted using convolutional neural network. back-propagation network to construct prediction models different collocations training. The area under receiver operator characteristic curve (AUC) was compare performance among models. Delong differences in AUC between pairwise. Results integrated clinical-morphological features, radiomic (CMRI) performed best models, achieved highest set, 1, which 0.886 (95% CI: 0.841-0.921), 0.830 (95%CI: 0.749-0.893) 0.879 0.712-0.968), In above three sets, CMRI other significant (all P < 0.05). Moreover, accuracy 96.88%. Conclusion feasible predict GGNs, is helpful refined management GGNs.