Enhancing Lung Nodule Classification Through Weighted Ensemble Framework

Rationale and Objectives : While Computed Tomography (CT) scans are often used to detect lung cancer, radiologists face challenges in differentiating and classifying nodules. By automating the analysis of imaging data, Deep Learning (DL) techniques result in better patient outcomes. Materials and Methods : To address this issue, we introduced an innovative Deep Convolutional Neural Network (DCNN)named RSANet, built upon the structure of ResNet50 for extracting features from lung images. However, an individual learner typically achieves unsatisfactory results. To address this issue, we suggest a weighted ensemble learner to improve the accuracy of nodule classification. The predictions of three models, including RSANet, modified ResNet50, and Xception models, were fused. The Lung Nodule Analysis 16 (LUNA16) dataset was used for performance evaluation. Results : The results show a substantial enhancement when employing the weighted ensemble technique. The method analysis shows a maximum accuracy of 97.61%, which has the highest value among all current research on the LUNA16. Conclusion : The enhancement of the proposed ensemble method demonstrates its robustness in ensemble learning, enabling it to achieve more optimal accuracy.