Abstract 300P
Background
Small cell lung cancer (SCLC) has the fastest growth rate and causes the earliest metastases of all lung cancer types. The seemingly normal parenchymal tissue surrounding the primary tumor may contain tiny foci, which increases the risk of distant metastasis (DM).
Methods
Intratumor and peritumor regions of SCLC were segmented from preprocessed CT images, and single region of interest (ROI) models and combined radiomics models, such as total tumor volume (GTV), peritumor volume (PTV), and combined total tumor volume and peritumor volume (GPTV), were built using machine learning algorithms. A combined radiometric-clinical model was constructed using selected clinical factors and externally validated at the Second Medical Center. An evaluation of the model’s performance was conducted by applying the Delong test, estimating the calibration curve, and determining the decision curve. A concordance index (CI) analysis was conducted on each cohort using univariate and multivariate tests.
Results
Our findings showed that the GPTV 0–3 mm radiomics model was accurately predicted by the work characteristics of participants in the training, internal validation, and external validation cohorts, with area under the curves (AUCs) of 0.943, 0.793, and 0.773, respectively. When combined with the clinical model, the AUCs for these cohorts were 0.945, 0.799, and 0.779, respectively.
Table 300PPredictive performance of models in the three cohorts
| AUC | Accuracy | Sensitivity | PPV | NPV | Recall | ||
| GTV | Training | 0.891 | 0.817 | 0.723 | 0.855 | 0.793 | 0.723 |
| Internal validation | 0.743 | 0.75 | 0.588 | 0.833 | 0.708 | 0.588 | |
| External validation | 0.729 | 0.667 | 0.565 | 0.565 | 0.73 | 0.565 | |
| GPTV0-3_FF | Training | 0.943 | 0.852 | 0.969 | 0.768 | 0.967 | 0.969 |
| Internal validation | 0.793 | 0.722 | 0.706 | 0.706 | 0.737 | 0.706 | |
| External validation | 0.773 | 0.75 | 0.609 | 0.7 | 0.775 | 0.609 | |
| GPTV0-3_FF+C | Training | 0.945 | 0.873 | 0.815 | 0.898 | 0.855 | 0.815 |
| Internal validation | 0.799 | 0.694 | 0.941 | 0.615 | 0.9 | 0.941 | |
| External validation | 0.779 | 0.733 | 0.739 | 0.63 | 0.818 | 0.739 |
Conclusions
CT imaging-derived radiomic features surrounding tumors in SCLC patients provide considerable predictive value for assessing the risk of DM.
Legal entity responsible for the study
The author.
Funding
Research Fund of Anhui Institute of translational medicine (2023zhyx-C40).
Disclosure
The author has declared no conflicts of interest.