<p>Chest radiography (CXR) remains one of the most widely used diagnostic tools for identifying lung diseases due to its availability, low cost, and non-invasive nature. Deep learning (DL) has proven to be an effective aid in screening and diagnosing lung diseases, thus reducing the workload of radiologists. This work introduces a novel diagnosis model designed for lung disease classification using CXR images, leveraging a five-stage approach to enhance diagnostic accuracy. Initially, preprocessing is performed through Gaussian filtering to reduce noise and improve image quality. Segmentation follows, utilizing a Layer Customized Patch-based U-Net (LCP-U-Net) model to accurately delineate lung regions. The extracted features include Improved Local Gabor Transitional Pattern (LGTrP), Median Binary Pattern (MBP), Statistical features, and Deep features, all contributing to a comprehensive feature representation. Data augmentation techniques are applied to expand the dataset and improve model generalization. The novelty of this work lies in the use of a hybrid deep learning architecture combining the Layer Customized Patch-based U-Net for segmentation and the Improved Squeeze Net-Pyramid Net for classification, ensuring high accuracy in diagnosing multiple lung diseases from CXR images. The objective was to create a highly accurate, user-friendly system for medical professionals to diagnose Tuberculosis, Pneumonia, and COVID-19, thereby improving healthcare delivery. The findings demonstrate the effectiveness of the proposed system, achieving a classification accuracy of 97% and a robust diagnostic framework, making it suitable for integration into clinical environments as a Software as a Medical Device (SaMD). The method can be integrated into existing medical imaging systems or deployed as a standalone application, supporting both diagnostic accuracy and accessibility in clinical environments.</p>