88P - Research of radiomics based on indeterminate lung nodules predicting prognosis of LARC patients

Background

With lung remains the second common site of colon-rectal cancer metastasis, it is still a challenge for early detection and finding. Therefore, we evaluate the CT-based radiomics of indeterminate lung nodules, predicting lung metastasis and prognosis in locally advanced rectal cancer patients.

Methods

A retrospective review for lung metastatic patients with colon-rectal cancer (CRC) and chest CT images were conducted. Radiomic prediction model of lung metastasis was trained by 114 patients with pathologically verified lung metastasis and 122 patients with benign lung nodules. We investigate the value of radiomics for identifying lung metastasis in 174 locally advanced rectal cancer(LARC) patients with follow-up information. Then, we conducted a Cox model and Kaplan-Meier curve analysis based on radiomics risk scores, and compare them to a clinical-pathological model for prognostic prediction. We use LASSO and linear regression to generated the radiomics model. C-index was used to assess model performance.

Results

For lung metastatic nodule identification in CRC patient, the C-index was 0.794(95%CI 0.784 -0.803) in the training set and 0.752(95%CI 0.728-0.776) in the validation set. In LARC patients, the C-index for lung metastatic identification was 0.771(95%CI 0.763-0.780). For prognostic prediction in LARC patients, ypTNM stage had a great influence on prognosis(Log-rank test P=0.003), and the C-index was 0.695 (95%CI 0.638-0.752). The C-index for nodules was 0.663(95%CI 0.575-0.751) with HR=1.148 (95% 1.050-1.256, P=0.003), and P<0.001 for Log-rank test. The combination of the ypTNM stage and nodule radiomics information has the C-index of 0.757 (95%CI 0.692-0.822), with P<0.001 for Log-rank test, which increase the performance of clinical prognostic prediction(P= 0.044).

Conclusions

Radiomics for nodules can determine lung metastasis in LARC patients. Lung nodules radiomics can provide information for prognostic analysis. The combination of lung nodules radiomics and ypTNM information increases the performance of prognostic prediction in LARC patients.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.