Enhanced genetic algorithm-optimized deep learning features for lung cancer classification

Lung cancer is a major global health problem and early detection is crucial to prevent serious health problems. Analyzing affected areas manually is difficult and requires skilled physicians. To help with this, AI tools have been developed for automated lung cancer detection. This paper proposes a novel method for detecting lung cancer using deep learning techniques. It uses two specific deep learning models, EfficientNet-b0 and InceptionResNet-V2, to compute important features from lung images. The EfficientNet-b0 model is adjusted by replacing certain layers to improve its performance in lung cancer data. After extracting the characteristics, a specially designed genetic algorithm helps select the most useful characteristics, reducing unnecessary features, and making the system more efficient. The improved GA reduces more than 50% of features without interfering with the recognition results. The proposed approach is validated on two publicly available datasets, the CT Scan Images for Lung Cancer Dataset and the IQ-OTH/NCCD Dataset, achieving a classification precision of 99.50% and 99.20%, respectively. Our approach achieved 0.5 to 2.5% higher accuracy in comparison to state-of-the-art methods while reducing the dimensionality of the features by more than 50%, without affecting the classification performance. The improved genetic algorithm smartly chooses key features, thus accelerating processing and lowering costs. It proves valuable in real-time medical applications and automated lung cancer detection, supporting early diagnosis and treatment planning. • Fine-tuned EfficientNet-b0 and InceptionResNet-V2 models for complementary deep feature extraction from lung CT images. • Proposed an improved Genetic Algorithm (GA) with adaptive weighting and redundancy penalty for effective feature selection. • Reduced feature dimensionality by more than 50% without sacrificing classification accuracy. • Achieved high classification accuracies of 99.50% and 99.20% on two public lung cancer datasets. • Outperformed state-of-the-art methods while reducing computational cost, making the framework suitable for real-time clinical use.