By combining thorough picture preprocessing with optimized feature extraction, this study offers a sophisticated approach for early lung cancer identification. It makes it possible to extract both multivariate static features and dynamic deep features from ResNet50 by improving lung image quality and segmentation to precisely designate regions of interest. By identifying intricate nonlinear and hidden patterns in lung nodules, our method improves prediction accuracy and facilitates prompt diagnosis and individualized treatment, thereby addressing the shortcomings of earlier research. We then combined features and fed to machine learning ensemble extreme boosting (XGBoost) algorithm by optimizing the hyperparameters using Bayesian optimization for improving the detection performance with and without feature selection methods. The proposed GLCM + ResNet50 method surpassed most existing methods, achieving a high accuracy of 97.62%, a Matthews Correlation Coefficient (MCC) of 94.94%, an F1‐Score of 94.03%, and an Area Under the Curve (AUC) of 0.9850 with top 400 hybrid features using Kruskal Wallis feature selection method. This hybrid approach, which effectively combines texture analysis with deep learning, demonstrates potential for robust and enhanced feature extraction. By capturing both local and global image features, this method leads to improved performance.
Maqbool et al. (Thu,) studied this question.