Optimized Feature Learning Using Convolutional Block Attention Module (CBAM) for Lung Cancer Detection: A Deep Learning Perspective

Lung cancer is one of the leading causes of cancer related deaths worldwide, which requires immediate and accurate detection for efficient treatment. Traditional methods like biopsies and CT scans are time taking and sujected to errors. This study proposes a custom deep learning model, where Convolutional Neural Network (CNN) along with CBAM - Convolutional Block Attention Module to enhance classification accuracy and feature extraction. Our model processes grayscale lung CT images resized to $256 \times 256$ pixels, leveraging CBAM to emphasize critical spatial and channel-wise features while filtering out irrelevant data. The IQ-OTHNCCD dataset, consisting of $\mathbf{1, 1 9 5}$ lung CT scans, was utilized for training and validation. To improve performance and class balance, prior processing techniques were used such as resizing, normalization, the synthetic minority oversampling technique- SMOTE, and augmentation. Adam optimizer was used in training the model with categorical cross-entropy loss for optimal convergence. Our model achieved $\mathbf{9 9. 6 4 \%}$ accuracy, $\mathbf{9 9. 9 9 \%}$ recall, $\mathbf{9 9. 9 2 \%}$ precision, and $\mathbf{9 9. 9 9 \%}$ F1-score. Confusion matrix analysis confirms high classification reliability, while graphical performance evaluations show stable learning convergence. The proposed CNN combined with CBAM model outclasses traditional and existing deep learning models, making it a highly reliable and effective solution for automated detection of lung cancer and clinical applications.