Delta Radiomics Model for the Prediction of Overall Survival and Local Recurrence in Small Cell Lung Cancer Patients After Chemotherapy

Rationale and Objectives To evaluate the validity of CT-based delta radiomics signatures in predicting overall survival (OS) and local recurrence (LR) in small cell lung cancer (SCLC) patients after chemotherapy. Materials and Methods Retrospectively enrolled 136 SCLC patients were split into training and testing cohorts. Radiomics features were extracted from CT images before, after the second, and the fourth cycle of chemotherapy. Delta radiomics features were obtained by calculating the net changes of features. Three radiomics signatures (R1, R2, and R3) and three delta radiomics signatures (R21, R31, and R32) were developed. The best signature was defined as the radiomics risk signature (RRS). The significant clinicoradiological factors and RRS of OS or LR were applied to build the combined model. RRS was also investigated in the subgroups based on stage and treatment regimens, respectively. Results Delta radiomics models presented improved performance. R32 signature demonstrated the highest C-indices in the training and testing cohorts, with C-indices of 0.850 and 0.834 in the OS arm, and 0.723 and 0.737 in the LR arm, respectively. The incremental performance was observed after the clinicoradiological characteristics integrated into the RRSOS, with C-indexes of 0.857 and 0.836, respectively. Furthermore, the stratified analysis also confirmed the ability of RRS based on the stage and treatment regimen subgroups in the OS and LR arms, respectively. Conclusion Delta radiomics signatures could improve the personalized prediction of OS and LR at the early stage of chemotherapy in SCLC patients. R32 signature performed the highest performance. To evaluate the validity of CT-based delta radiomics signatures in predicting overall survival (OS) and local recurrence (LR) in small cell lung cancer (SCLC) patients after chemotherapy. Retrospectively enrolled 136 SCLC patients were split into training and testing cohorts. Radiomics features were extracted from CT images before, after the second, and the fourth cycle of chemotherapy. Delta radiomics features were obtained by calculating the net changes of features. Three radiomics signatures (R1, R2, and R3) and three delta radiomics signatures (R21, R31, and R32) were developed. The best signature was defined as the radiomics risk signature (RRS). The significant clinicoradiological factors and RRS of OS or LR were applied to build the combined model. RRS was also investigated in the subgroups based on stage and treatment regimens, respectively. Delta radiomics models presented improved performance. R32 signature demonstrated the highest C-indices in the training and testing cohorts, with C-indices of 0.850 and 0.834 in the OS arm, and 0.723 and 0.737 in the LR arm, respectively. The incremental performance was observed after the clinicoradiological characteristics integrated into the RRSOS, with C-indexes of 0.857 and 0.836, respectively. Furthermore, the stratified analysis also confirmed the ability of RRS based on the stage and treatment regimen subgroups in the OS and LR arms, respectively. Delta radiomics signatures could improve the personalized prediction of OS and LR at the early stage of chemotherapy in SCLC patients. R32 signature performed the highest performance.