Computer assisted diagnosis of respiratory diseases, such as Pneumonia, COVID-19, and Tuberculosis, based on chest X-ray (CXR) images, is of crucial importance in clinical diagnostics. The suggested approach employs EfficientNet-B0 together with a Convolutional Block Attention Module (CBAM) in order to advance feature extraction capacity and increase classification efficiency. Data augmentation strategies are employed with the aim to advance model generalizability and, therefore, to enable high-prediction power across datasets. With the aid of attention mechanisms, the hybrid model is effective in making appropriate regions within CXR images, thereby improving performance and reducing rates of misclassification. Experimental outcomes show that the suggested methodology performs better compared to conventional deep learning models and hence its usability in reliable computer assisted disease diagnosis in the clinic.