<sup>18</sup>F-FDG PET/CT and receptor-positive circulating tumor cells-based machine learning model for predicting poorly differentiated lung adenocarcinoma.
Authors
Affiliations (7)
Affiliations (7)
- Department of Nuclear Medicine, Shanghai Pulmonary Hospital, School of Medicine, Tongji University, Shanghai, China.
- Department of Radiology, Shandong Cancer Hospital and Institute, Shandong First Medical University and Shandong Academy of Medical Sciences, Jinan 250117, Shandong, China.
- Department of Ultrasound, Shanghai Pulmonary Hospital, School of Medicine, Tongji University, Shanghai, China.
- Department of PET/CT Center, Shandong Cancer Hospital and Institute, Shandong First Medical University and Shandong Academy of Medical Sciences, Jinan 250117, Shandong, China.
- Department of Radiology, Shandong Cancer Hospital and Institute, Shandong First Medical University and Shandong Academy of Medical Sciences, Jinan 250117, Shandong, China. Electronic address: [email protected].
- Department of Nuclear Medicine, Shanghai Pulmonary Hospital, School of Medicine, Tongji University, Shanghai, China. Electronic address: [email protected].
- Department of Ultrasound, Shanghai Pulmonary Hospital, School of Medicine, Tongji University, Shanghai, China. Electronic address: [email protected].
Abstract
This study aimed to develop and validate a machine learning model that integrates radiomic features from 2-[<sup>18</sup>F]fluoro-2-deoxy-D-glucose (<sup>18</sup>F-FDG) positron emission tomography/computed tomography (PET/CT) with folate receptor-positive circulating tumor cells (FR<sup>+</sup>-CTCs) for the preoperative prediction of tumor differentiation grade, as defined by the International Association for the Study of Lung Cancer (IASLC) grading system, in patients with clinical stage IA lung adenocarcinoma (ADC). This retrospective study enrolled a total of 1797 patients from two medical centers. Pathological evaluation identified 1008 cases as poorly differentiated tumors (PDT) and 789 cases as non-poorly differentiated tumors (n-PDT). Three kinds of models were constructed, including the clinical, radiomics, and combined model. The combined model was established using 5 machine learning algorithms. Model performance was assessed by the area under the receiver operating characteristic curve (AUC), with Shapley Additive Explanations (SHAP) employed for interpretability. The light gradient boosting machine (LightGBM) combined model, incorporating FR<sup>+</sup>-CTCs and 11 radiomic features, demonstrated superior predictive performance compared to other models, with AUCs of 0.960, 0.906, and 0.902 in the training, internal validation, and external validation sets. Additionally, this model exhibited favorable calibration and high net benefit. Progression-free survival (PFS) was significantly different between the PDT and n-PDT patients, and between the high-risk and low-risk patients stratified by this model (both p < 0.001). The LightGBM combined model effectively predicts tumor differentiation grade in patients with clinical stage IA lung ADC and can be used as a tool for risk stratification of these patients.