X-ray imaging is one commonly employed method among the range of diagnostic tools available for detecting lung tuberculosis. Pulmonary tuberculosis is often detected and screened through chest X-rays (CXR). Nevertheless, the examination of chest radio-graphs by skilled physicians to identify Tuberculosis (TB) in clinical settings can be time-consuming and prone to subjective inconsistencies. Medical experts analyzing medical data face substantial subjectivity and variation in image quality. Therefore, there is growing interested in the healthcare sector to leverage machine learning to improve the accuracy and efficiency of medical data analysis. Considering this matter, Digital Pathology (DP) and Algorithmic machine-assisted diagnosis (AMAD) systems have the potential to significantly aid in the large-scale examination of pulmonary tuberculosis through the analysis of chest X-ray image features. We propose OrthoSNet: a network architecture that learns image contrast translation invariant features. We have made use of the Tuberculosis Dataset for the development and substantiation of the model. Our experiment validates the practicality and utility of our alternative approach, With a significant improvement in accuracy. The proposed model predicts a superior accuracy of 99 % with a sample size of 1400 images.