DARNet: Deep Attention Module and Residual Block-Based Lung and Colon Cancer Diagnosis Network

Accurate and efficient lung and colon cancer classification is vital for early detection and treatment planning. Traditional methods require manual effort and expert analysis, leading researchers to explore deep learning models. However, deep learning-based lung and colon cancer classification models face challenges such as generalization, overfitting, gradient vanishing, and hyperparameter tuning. To overcome these challenges, we propose an efficient Deep Attention module and a Residual block-based lung and colon cancer classification Network (DARNet). It comprises three key components such as residual blocks, attention modules, and fully connected layers. Residual blocks (RBs) are utilized to refine the DARNet's ability to learn and capture residual information which allows DARNet to perceive complex patterns and improve accuracy. Attention module (AM) enhances feature extraction and captures useful information in the input data. Finally, to achieve better generalization performance, we employ Bayesian Optimization (BO) to fine-tune the hyperparameters of DARNet. Extensive experimental results indicate that the proposed BO-based DARNet achieved superior performance over competitive models on benchmark lung and colon cancer datasets, with a median accuracy of 98.86% and lower variance.