AI-derived 3D body composition predicts prognosis of chemoimmunotherapy in small cell lung cancer: a multicenter study.
Authors
Affiliations (9)
Affiliations (9)
- Department of Radiology, Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China.
- Hubei Key Laboratory of Molecular Imaging, Wuhan, China.
- Hubei Provincial Clinical Research Center for Precision Radiology & Interventional Medicine, Wuhan, China.
- Department of Thoracic Surgery, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China.
- Department of Radiotherapy, Fujian Medical University, Fuzhou, China.
- Medical School of Chinese PLA, Beijing, China.
- Xiangyang No.1 People's Hospital, Hubei University of Medicine, Xiangyang, China.
- Institute of Research and Clinical Innovations, Neusoft Medical Systems, Shanghai, China.
- Department of Radiology, Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China [email protected] [email protected] [email protected].
Abstract
Small cell lung cancer (SCLC) has a poor prognosis despite standard first-line chemoimmunotherapy. Reliable, non-invasive biomarkers for risk stratification are urgently needed. Three-dimensional (3D) body composition (BC) analysis from chest CT scans offers a promising approach to assess systemic metabolic status. In this multicenter study, we analyzed patients treated with first-line chemoimmunotherapy across five institutions. Using a deep learning model, 3D BC parameters were automatically extracted from baseline chest CTs. We evaluated their associations with overall survival (OS) and progression-free survival using Kaplan-Meier and Cox regression analyses. A total of 101 machine learning model configurations were evaluated to develop predictive models, and the SHapley Additive exPlanations method was applied to enhance model interpretability. Additionally, we investigated the association between BC parameters and CD8<sup>+</sup> T-cell infiltration in the tumor microenvironment. Among 575 patients (training: 161; internal validation: 69; external validation: 345), all four 3D BC indices significantly stratified OS. Multivariable Cox analysis showed that higher skeletal muscle volume index was independently associated with better OS (HR=0.956, 95% CI 0.928 to 0.985, p=0.003), whereas higher visceral fat volume index was associated with worse OS (HR=1.017, 95% CI 1.002 to 1.032, p=0.026). 3D BC provided more pronounced prognostic stratification than conventional two-dimensional measures and body mass index. The selected linear discriminant analysis model achieved area under the curves of 0.756, 0.781, and 0.783 in the training, internal validation, and external validation cohorts, respectively. In an exploratory subset of 35 patients, higher skeletal muscle volume index was associated with increased intratumoral CD8+T cell density. Higher skeletal muscle volume index was associated with better OS, whereas higher visceral fat volume index was associated with worse OS. Automated 3D BC may provide complementary non-invasive prognostic information in patients with SCLC receiving first-line chemoimmunotherapy.