A novel dense-net deep neural network with enhanced feature selection method for classification of different stages of tuberculosis using chest X-ray images

Chest X-ray radiography (CXR) is a widely used medical imaging technique, essential for diagnosing potentially fatal diseases. Radiologists typically analyze CXR images to identify conditions, but the visual similarities among various lung diseases can lead to misdiagnoses. This challenge has sparked significant interest in the medical imaging field, particularly in the development of computer-aided diagnosis (CAD) systems to improve the accuracy of detecting lung diseases in CXR images. This research introduces CBAMWDnet, an innovative deep learning model designed for tuberculosis (TB) diagnosis using chest X-ray (CXR) images. The model leverages the Wide Dense Net (WDnet) architecture and the Convolutional Block Attention Module (CBAM) to effectively extract spatial and contextual information. Prior to applying the model, an enhanced Gabor filter is used as part of the pre-processing strategy. This research introduces CBAMWDnet, an innovative deep learning model designed for tuberculosis (TB) diagnosis using chest X-ray (CXR) images. The model leverages the Wide Dense Net (WDnet) architecture and the Convolutional Block Attention Module (CBAM) to effectively extract spatial and contextual information. Prior to applying the model, an enhanced Gabor filter is used as part of the pre-processing strategy. The proposed model is evaluated using the Shenzhen, Montgomery, and Indiana datasets, which are freely accessible and widely recognized in the field. Among these, the Shenzhen dataset delivers the best results. The proposed method outperforms all existing approaches across multiple datasets. Our system effectively categorizes tuberculosis into four distinct phases: normal, primary infection, latent infection, and active illness. This classification enhances the accuracy and reliability of TB diagnosis, demonstrating the superior performance of our approach compared to other methods currently in use.