Value of 18F-FDG PET/CT radiomics for predicting KRAS gene mutations in non-small cell lung cancer / 中华核医学与分子影像杂志

ABSTRACT

Objective:To assess the predictive efficacy of 18F-FDG PET/CT-based radiomics models for the mutation status of Kirsten rats sarcoma viral oncogene homolog (KRAS) in patients with non-small cell lung cancer (NSCLC). Methods:From January 2016 to January 2021, the 18F-FDG PET/CT images and KRAS testing of 258 NSCLC patients (180 males, 78 females; age 33-91 years) in the First Affiliated Hospital of the Air Force Military Medical University were retrospectively analyzed. Patients were randomly divided into training set ( n=180) and validation set ( n=78) in the ratio of 7∶3. Tumor lesions on PET and CT images were drawn respectively, and the radiomics features of PET and CT lesions were extracted. The radiomics features were screened by least absolute shrinkage and selection operator (LASSO). CT radiomics score (RS) model, PET/CT RS model and composite models of PET/CT RS combined with screened clinical information were eventually developed. ROC curves were used to assess the predictive efficacy of these models. Results:The CT RS model included 4 radiomics features and the PET/CT RS model included 4 CT radiomics features and 8 PET radiomics features. The CT RS model and the PET/CT RS model both had significant differences in RS between KRAS mutant and wild-type patients in the training set and validation set ( z values from -8.30 to -4.10, all P<0.001). In predicting KRAS mutations, the composite model of PET/CT RS combined with age showed AUCs of 0.879 and 0.852 in the training and validation sets respectively, which were higher than those of the CT RS model (0.813 and 0.770) and the PET/CT RS model (0.858 and 0.834). The accuracy of the composite model of PET/CT RS combined with age were 81.67%(147/180) and 79.49%(62/78) in the training set and validation set respectively, which were also higher than those of the CT RS model (75.00%(135/180) and 74.36%(58/78)) and the PET/CT RS model (78.89%(142/180) and 78.21%(61/78)). Conclusion:Models based on radiomics features can predict KRAS gene mutation status, and the composite model combining PET/CT RS and age can improve the prediction performance.