Construction of a radiogenomics predictive model for KRAS mutation status in patients with non-small cell lung cancer.

BACKGROUND

Non-small cell lung cancer (NSCLC) represents a significant portion of lung cancer cases globally, with kirsten rats arcomaviral oncogene homolog (KRAS) mutations being a critical factor in its pathogenesis. Predicting KRAS mutation status is crucial for guiding targeted therapies and improving patient outcomes. This study aimed to develop and validate a differential evolution optimized artificial neural network (DE-ANN) model that integrates positron emission tomography/computed tomography (PET/CT) radiomics and genomics data for predicting KRAS mutation status in NSCLC patients, showcasing the potential of multi-omics integration in precision oncology.

METHODS

The study utilized PET/CT radiomics features and genomics data from public databases using least absolute shrinkage and selection operator (LASSO) regression and support vector machine-recursive feature elimination (SVM-RFE) to identify key predictive features. The DE-ANN model was optimized using differential evolution algorithms and validated internally using Bootstrap resampling to assess its predictive performance.

RESULTS

The DE-ANN model demonstrated superior predictive accuracy with an area under the curve (AUC) of 0.909 [95% confidence interval (CI): 0.882-0.937], outperforming traditional artificial neural network (ANN) models (AUC =0.819, 95% CI: 0.778-0.860). Key features identified included significant radiomics signatures and gene markers, with the model showing enhanced convergence rates and robust internal validation outcomes. The model's calibration and decision curve analyses further confirmed its clinical applicability and potential for improving personalized treatment strategies in NSCLC.

CONCLUSIONS

The DE-ANN model represents a significant advancement in the predictive modeling of KRAS mutation status in NSCLC, leveraging the synergy between radiomics and genomic data. Its high predictive accuracy and methodological robustness highlight the model's potential as a tool in precision oncology, warranting further external validation and exploration in other cancer types.