INTRODUCTION
Non-small cell lung cancer (NSCLC), particularly adenocarcinoma, is a molecularly diverse disease where EGFR mutations serve as critical biomarkers for guiding targeted therapies, especially tyrosine kinase inhibitors (TKIs). In clinical settings, the rapid and non-invasive identification of EGFR mutation status is essential for initiating precision treatment. This study addresses that need by developing a nomogram combining the most informative clinical, CT, and radiomic features to predict EGFR mutation status. By using real-world retrospective data from 521 histologically confirmed NSCLC adenocarcinoma patients, the study creates a reliable tool to support early therapeutic decisions. The research demonstrates how integrating multi-modal data and machine learning can improve clinical decision-making without depending solely on invasive biopsies.
DATA ACQUISITION AND STUDY POPULATION
The study involved 521 NSCLC adenocarcinoma patients who underwent CT imaging along with either surgical resection or biopsy for EGFR mutation testing. This real-world, retrospective dataset ensures that the results are clinically applicable and mirror actual diagnostic scenarios. Patient demographics and medical records were examined to extract the most relevant clinical data, and imaging data were processed to extract detailed CT and radiomic features. This diverse dataset forms the foundation for developing and validating the predictive models and reflects the heterogeneity typically encountered in clinical practice.
FEATURE EXTRACTION AND SELECTION STRATEGY
Three major feature types were collected: clinical variables (e.g., age, sex), CT-based morphological features, and radiomic features derived from regions of interest (ROIs) in the CT scans. Radiomic analysis allowed the quantification of tumor heterogeneity and textural complexity beyond visual assessment. A feature preselection process was employed to identify 11 key signatures (2 clinical, 2 CT-based, and 7 radiomic), which proved to have the highest predictive value. This step minimized redundancy, reduced dimensionality, and enhanced the interpretability and generalizability of the final model.
MACHINE LEARNING MODEL DEVELOPMENT
To evaluate the predictive power of different feature combinations, five Random Forest classifiers were trained on various datasets. These datasets ranged from raw unfiltered data to optimized subsets with only the most relevant features. Among all configurations, the model trained exclusively on the preselected 11 features showed the highest performance metrics. Random Forest, a robust ensemble-based learning algorithm, was selected for its ability to handle feature heterogeneity and complex interactions, making it well-suited for integrating clinical and imaging data.
RESULTS AND PERFORMANCE METRICS
The optimized Random Forest model yielded superior predictive performance across all key evaluation metrics. It achieved an AUC of 0.88, precision of 0.90, recall of 0.94, F1-score of 0.91, and overall accuracy of 0.87. Both macro- and micro-average scores exceeded 0.89, underscoring the model’s strong classification capabilities for distinguishing EGFR-mutant from wild-type NSCLC cases. These results indicate that the selected multi-modal features effectively capture the biological and morphological signatures associated with EGFR mutation, validating the approach for clinical use.
CLINICAL IMPLICATIONS AND CONCLUSION
The developed nomogram offers a robust, non-invasive tool for the early prediction of EGFR mutation status in NSCLC adenocarcinoma patients. It supports precision treatment planning by enabling rapid identification of candidates for TKI therapy without the delay or risk associated with biopsy-based molecular testing. This approach bridges imaging and molecular biology through machine learning, promoting personalized care. Future research may involve prospective validation and integration into clinical workflows to maximize real-world utility and improve outcomes in lung cancer management.
📢 Visit: https://infectious-diseases-conferences.pencis.com
🏆 Nominate Now: https://infectious-diseases-conferences.pencis.com/award-nomination/?ecategory=Awards&rcategory=Awardee
📝 Registration Page: https://infectious-diseases-conferences.pencis.com/award-registration/
📧 Contact us: infectioussupport@pencis.com
Hashtags:
#EGFRMutation, #NSCLC, #LungCancer, #Radiomics, #MachineLearning, #CTImaging, #PrecisionMedicine, #TyrosineKinaseInhibitors, #CancerDiagnostics, #NonInvasiveTesting, #ClinicalAI, #RandomForest, #MedicalAI, #CancerPrediction, #DeepRadiomics, #EGFRStatus, #OncologyResearch, #NomogramModel, #ImageAnalysis, #PencisConference,
No comments:
Post a Comment