Residual-SwishNet: a deep learning-based approach for reliable lung cancer classification

Lung cancer remains one of the primary causes of cancer-related deaths globally, emphasizing the urgent need for accurate and early diagnosis to improve patient outcomes. However, existing computer-aided detection systems often struggle with suboptimal feature extraction, low classification accuracy, and limited generalizability across datasets.

To address these challenges, we propose a deep learning approach named Residual-SwishNet, explicitly designed for the lung cancer classification task. More specifically, we modified the ResNet50 framework by replacing the conventional ReLU activation function with Swish during the feature engineering phase. Further, we integrate three additional dense layers before the classification module to obtain an enriched feature representation. Lastly, we employ a Softmax output layer with Cross-Entropy Loss to tackle the class-imbalance issue.

The approach was rigorously evaluated on 2 publicly accessible datasets, named LUNA16 and IQOTH/NCCD, using precision, recall, F1-score, and accuracy as performance metrics. Experimental results demonstrate the superiority of our technique, achieving classification accuracies of 99.60% and 99.11% on the LUNA16 and IQ-OTH/NCCD datasets.

Our approach has significantly outperformed existing state-of-the-art techniques. These findings highlight the potential of the proposed model as a robust and reliable tool for lung cancer diagnosis.