Enhanced Lung Cancer Detection Using a Hybrid AlexNet-CNN and Handcrafted Feature Fusion with mRMR Feature Selection and SSA-GWO Optimized Segmentation

Abstract

The recognition and diagnosis of lung cancer is a major component in clinical decision-making with early and accurate detection of the cancerous condition being fundamental to improving patient outcomes. The aim of the paper is to present a new variant of chest cancer detection based on deep learning with the incorporation of hand-crafted radiomic features into the processing pipeline of multi-stage processing. Hybrid feature extraction method implements deep learning of the AlexNet CNN together with what can be considered as handcrafted radiomic features, e.g., HOG, Gabor filters, and Wavelets transforms, allowing to provide more detailed representation of tumor features. Also, Minimum Redundancy Maximum Relevance (mRMR) method is applied to feature selection by keeping the most relevant and non-redundant features and eliminating noise to boost accuracy of classification. The support vector machines (SVM), as well as an ensemble classification model, specifically the Random Forest (RF) algorithm, introduces both robustness and accuracy in this case, since they effectively deal with high-dimensional and complex data. The approach is used to overcome the problems with the classification of lung cancer as it raises the accuracy, efficiency, and practicability of the model in clinical practice. The given methodology is proven to present a dependable and effective solution that may be used in the real-time clinical settings to detect lung cancer early and accurately.