Automatic segmentation and CT-based deep learning radiomics nomogram for predicting overall survival in patients with small cell lung cancer: A multicenter cohort study.
Authors
Affiliations (6)
Affiliations (6)
- Department of Radiology, The First Affiliated Hospital of Anhui Medical University, Hefei 230031, China.
- Department of Radiology, The Third Affiliated Hospital of Anhui Medical University, Hefei, 230061, China.
- Department of Radiology, The First Affiliated Hospital of USTC, Division of Life Sciences and Medicine, University of Science and Technology of China, Hefei 230001, China; Department of Radiology, Anhui Provincial Cancer Hospital, Hefei 230031, China.
- Department of Radiology, The Affiliated Bozhou Hospital of Anhui Medical University, Bozhou 236000, China.
- Department of Radiology, The Affiliated Bozhou Hospital of Anhui Medical University, Bozhou 236000, China. Electronic address: [email protected].
- Department of Radiology, The First Affiliated Hospital of Anhui Medical University, Hefei 230031, China. Electronic address: [email protected].
Abstract
Accurate prediction of overall survival (OS) in patients with small cell lung cancer (SCLC) is crucial for personalized treatment. This study aimed to create a three-dimensional (3D) automatic segmentation model for identifying SCLC lesions in computed tomography (CT) images. Moreover, we sought to develop and validate a deep learning radiomics nomogram (DLRN) utilizing pretreatment CT images to predict OS in SCLC patients. A total of 1061 SCLC patients from four hospitals in China were retrospectively enrolled. A 3D automatic segmentation model for SCLC lesions was developed using the nnU-Net neural network. Radiomics and deep learning features were extracted from the 3D tumor volume on the basis of pretreatment CT images of arterial phase (AP) and venous phase (VP). Subsequently, the radiomics score (Rad-score) and deep learning score (DL-score) were constructed. An integrated DLRN was constructed by combining the Rad-score and DL-score, followed by assessments of its discrimination, calibration, reclassification, and clinical usefulness. The Dice similarity coefficients of the 3D automatic segmentation model on the AP and VP image test sets were 0.878 and 0.872, respectively. The DLRN showed satisfactory predictive performance for OS and yielded concordance indices of 0.892, 0.873, and 0.872 for the internal validation cohort, external validation cohort 1, and external validation cohort 2, respectively, with good calibration in all cohorts. Furthermore, the DLRN outperformed the single model and significantly outperformed the clinical nomogram (all P < 0.05). However, the addition of clinical factors did not improve the predictive efficacy of the DLRN on the basis of the net reclassification improvement and integrated discrimination improvement (all P > 0.05). The 3D automated segmentation model performed highly accurate segmentation of SCLC lesions, and a CT-based DLRN exhibited strong potential in predicting clinical outcomes for SCLC patients, potentially offering valuable insights for individualized therapy.