A novel Convolutional Shuffle Attention Xtreme Gradient Boost Network for improved lung cancer detection using computed tomography images.

Lung cancer is a severe and life-threatening type of cancer that originates in the lung tissues. Computed Tomography (CT) image emerges as the primary diagnostic tool for identifying lung cancer. However, manual interpretation of CT images necessitates the development of automated techniques. Hence, this article proposes a hybrid model, Convolutional Shuffle Attention Xtreme Gradient Boost Network (SA-XGBNet) for detecting lung cancer through CT images. The SA-XGBNet is the integration of Shuffle Attention Network (SA-Net), Convolutional Xtreme Gradient Boost (ConvXGB) and Fractional Calculus (FC). Initially, CT input images are composed from a database and filtered by Kolmogorov-Wiener Filter. Next, the pre-processed images undergo segmentation, performed using a Dual Attention Network (DA-Net) to isolate the lung nodule region. The image augmentation models, such as shearing, color jittering, resizing, flipping, and rotating, are applied. Following that, different features, including shape-based, intensity-based, Histogram of Oriented Local Binary Pattern Descriptor (HOLBP) with entropy and texture-based features are extracted. Finally, lung cancer is detected by SA-XGBNet. At 90 % training data, SA-XGBNet acquired 92.975 % accuracy, 94.977 % True Positive Rate (TPR), and 90.866 % True Negative Rate (TNR) using the Lung Image Database Consortium Image Collection and Image Database Resource Initiative (LIDC-IDRI).