Predictive models of epidermal growth factor receptor mutation in lung adenocarcinoma using PET/CT-based radiomics features.

Abstract

Background: Lung adenocarcinoma (LAC) comprises a substantial subset of non-small cell lung cancer (NSCLC) diagnoses, where epidermal growth factor receptor (EGFR) mutations play a pivotal role as indicators for therapeutic intervention with targeted agents. The emerging field of radiomics, which involves the extraction of numerous quantitative attributes from medical imaging, when coupled with positron emission tomography/ computed tomography (PET/CT) technology, has demonstrated promise in the prognostication of EGFR mutation status. The objective of this investigation is to construct and validate predictive models for EGFR mutations in LAC by leveraging PET/CT-derived radiomics features, thereby refining diagnostic precision and facilitating tailored treatment strategies.

Purpose: The aim of this study was to develop a non-invasive radiomics model based on PET/CT with excellent performance for predicting the EGFR mutation status in LAC. Thus, it can provide the basis for the individualized treatment decision of patients.

Methods: Positron emission tomography (PET), computed tomography (CT), clinical and pathological data of 112 patients with LAC admitted to our hospital from January 2019 to June 2023 were retrospectively analyzed. This research cohort encompassed 54 LAC patients with EGFR wild type and 58 LAC patients with EGFR mutated type. The participants were randomly assigned to the training group (n = 78) and the validation group (n = 34) in a 7:3 ratio. A sum of 3562 radiomics attributes were derived from PET/CT scans. The minimal absolute shrinkage and selection operator method was employed to identify 13 notable features. Based on these characteristics, support vector machine (SVM), gradient boosting decision tree (GBDT), random forest (RF) and extreme gradient boosting (XGBOOST) were constructed. The forecasting effectiveness of the model was assessed using the area under the receiver operating characteristic (ROC) Curve, the DeLong test, and decision curve analysis (DCA).

Results: SVM performance in PET/CT radiomics model was higher than that of other machine learning models (training group areas under the curve [AUC] of 0.916 and validation group AUC of 0.945, respectively). The integration of radiomics and clinical data did not yield a superior predictive performance compared to the radiomics model alone in terms of estimating EGFR mutation status (AUC: 0.916 vs. 0.921, 0.945 vs. 0.955, p> 0.05, in both the training and validation groups).

Conclusions: The SVM model has emerged as a commendable non-invasive technique, showing high precision and dependability in forecasting EGFR mutation statuses in individuals with LAC. The radiomics model derived from PET/CT scans holds promise as a prognostic indicator of EGFR mutations in LAC, offering a valuable tool that could refine personalized therapeutic strategies and ultimately enhance the prognosis for LAC patients.