Objective:

To explore the predict

ive value of computed tomography (CT) radiomics combined with clinical prediction model for distant metastasis of non-small cell lung cancer (NSCLC).

Methods:

Retrospective analysis of 140 lung cancer patients (74 without distant metastasis and 66 with distant metastasis)

two regions of interest (ROI)

including the peritumoral region and the tumor parenchyma region

were delineated from the pre-treatment CT images of each patient. The radiomics features were extracted from the two ROI

respectively

and the radiomics score (RS) was calculated. Consistency of feature delineation was checked by calculating the intraclass correlation coefficient (ICC). All data were grouped

with training and validation cohorts of 97 and 43

respectively. Statistical differences in the characteristics of the metastatic and non-metastatic groups were evaluated using the chi- square test or Fishers exact test. Four prediction models were constructed

namely tumor radiomics model

including peritumoral radiomics model and two comprehensive models combined with clinical features respectively. All model performance was quantified by the area under curve (AUC) of receiver operating characteristic (ROC) curve

the diagnostic ability of different models was tested by DeLong test

and a nomogram was constructed containing a comprehensive model of peritumoral radiomics combined with clinical features

and the calibration and discriminative ability of the nomogram was evaluated.

Results:

Nine radiomics signatures were screened in the tumor radiomics model alone

and thirteen radiomics signatures were screened in the including peritumoral radiomics model. The clinical factor carcinoembryonic antigen (CEA) had statistical significance in the prediction of distant metastasis of NSCLC (

P

＜0.05). The AUC values of the radiomics model alone were 0.779 and 0.854. Combining clinical factor

the combined models were 0.795 and 0.858

respectively.

Conclusion:

The model constructed by imaging c

haracteristics combined with clinical factor can be used to predict the DM of NSCLC

and the including peritumoral model can improve the ability to predict distant metastasis of NSCLC.