<p>Lung image segmentation can precisely identify and delineate lung features and potential anomalies in medical imaging, including X-ray images. This process is vital for the automated identification of tuberculosis disease. However, it struggles to accurately delineate small or complex lung structures, leading to the misidentification of subtle anomalies in X-ray images. In this manuscript, improving lung image segmentation for automated tuberculosis disease detection in medical systems (APDD-BEMS-RICCNN) is discussed. Initially, input images are sourced from the Montgomery County X-ray dataset. A Constrained Normalized Subband and Adaptive Filter (CNSAF) is used for pre-processing in order to standardize, resize, and normalize these images. Subsequently, the images undergo lung parenchyma segmentation using Dual Information Enhanced Multi-view Attributed Graph Clustering (DIEMAGC). This clustering-based method extracts structural and attribute-level features from multiple views of the image graph, effectively isolating lung boundaries while preserving fine anatomical details. The segmented lung regions, free from irrelevant background and non-tuberculosis structures, are then fed into a Rotation-Invariant Coordinate Convolutional Neural Network (RICCNN), integrated with an Attention-Enhanced Co-Interactive Fusion Network (AECIF-Net), to detect and classify tuberculosis into Normal and Abnormal categories. The performance measures of the proposed APDD-BEMS-RICCNN technique contain improved accuracy (99.14%), precision(98.49%), and Intersection over Union (IoU) (98.25%) when compared to the existing methods, like a deep learning architecture for multi-class lung diseases classification using chest X-ray (CXR)images (MCLDC-CXRI-CNN), novel hybrid deep learning method for early detection of lung cancer using neural networks (EDLC-3D-CNN), and a lung segmentation in chest X-ray image using multi-interaction feature fusion network (LS-CXRI-MFFN) respectively.</p>