PET/CT Radiomic Sequencer for Prediction of EGFR and KRAS Mutation Status in NSCLC Patients

TL;DR

This study developed PET/CT radiomic models using machine learning to non-invasively predict EGFR and KRAS mutation status in NSCLC patients, achieving up to 0.75 AUC performance.

Contribution

The paper introduces a novel radiomic approach combining multiple features and machine learning methods for mutation prediction in NSCLC, outperforming conventional PET parameters.

Findings

Radiomic models achieved AUC up to 0.75 for mutation prediction.

Combining wavelet, Laplacian of Gaussian, and discretization features improved accuracy.

Radiomics outperformed conventional PET features in predictive performance.

Abstract

The aim of this study was to develop radiomic models using PET/CT radiomic features with different machine learning approaches for finding best predictive epidermal growth factor receptor (EGFR) and Kirsten rat sarcoma viral oncogene (KRAS) mutation status. Patients images including PET and CT [diagnostic (CTD) and low dose CT (CTA)] were pre-processed using wavelet (WAV), Laplacian of Gaussian (LOG) and 64 bin discretization (BIN) (alone or in combinations) and several features from images were extracted. The prediction performance of model was checked using the area under the receiver operator characteristic (ROC) curve (AUC). Results showed a wide range of radiomic model AUC performances up to 0.75 in prediction of EGFR and KRAS mutation status. Combination of K-Best and variance threshold feature selector with logistic regression (LREG) classifier in diagnostic CT scan led to the best performance in EGFR (CTD-BIN+B-KB+LREG, AUC: mean 0.75 sd 0.10) and KRAS (CTD-BIN-LOG-WAV+B-VT+LREG, AUC: mean 0.75 sd 0.07) respectively. Additionally, incorporating PET, kept AUC values at ~0.74. When considering conventional features only, highest predictive performance was achieved by PET SUVpeak (AUC: 0.69) for EGFR and by PET MTV (AUC: 0.55) for KRAS. In comparison with conventional PET parameters such as standard uptake value, radiomic models were found as more predictive. Our findings demonstrated that non-invasive and reliable radiomics analysis can be successfully used to predict EGFR and KRAS mutation status in NSCLC patients.