Tuberculosis (TB) remains a significant global health challenge, with a high rate of infection and mortality. Chest X-ray (CXR) imaging is a common diagnostic tool for TB, but its effectiveness depends heavily on the expertise of the radiologist. This study explores the impact of image preprocessing on the performance of deep learning models in TB detection from CXR images, evaluating whether the computational cost of preprocessing is justified compared to using non-preprocessed images. A combination of all these preprocessing techniques was applied on the dataset, including Contrast Limited Adaptive Histogram Equalization (CLAHE), wavelet transform, gamma correction, and histogram equalization, as provided by the dataset itself. The results indicate that preprocessing enhanced the accuracy of the ResNet50 model significantly, achieving 99% accuracy compared to 94% on raw images. However, for MobileNet and the custom model, the improvement was marginal, suggesting that these models can perform adequately without extensive preprocessing. This finding highlights the potential for implementing deep learning models in low-resource settings where computational capabilities are limited. The study underscores the importance of selecting appropriate preprocessing techniques and neural network architectures to optimize TB detection accuracy in diverse clinical environments.