Integrating technology such as artificial intelligence in the medical field may contribute significantly to the early diagnosis of diseases. This study is conducted to achieve two goals: The first is to use three advanced deep learning models to identify tuberculosis and COVID-19 diseases from healthy cases by processing a chest X-ray image dataset. The second goal is to compare the three models and identify the most generalizable one based on accuracy and computational efficiency. The proposed models were applied to a large-scale chest X-ray of 6375 images. The findings show that InceptionV3 outperforms the CNNs and ResNet-50 algorithms by achieving high metrics values (Accuracy = 0.9636, Precision = 0.9752, Recall = 0.9554, and F1-Score = 0.9651). Furthermore, the results show that training neural networks across a wide range of epochs improves the model predictions of new data. The findings demonstrate InceptionV3's capability of providing a robust and automated method for classifying chest X-ray images. Apart from accuracy assessment, this computational analysis depicted marked variations in efficiency amongst the three models. The CNN architecture achieved the shortest training time of ≈12.4 minutes and attained the fastest inference speed of ≈4.1 ms per image, whereas ResNet-50 and InceptionV3 required longer training and inference times. This starkly indicates that there is a clear trade-off between computational cost and predictive performance across the evaluated architectures.