<p>Pneumonia remains one of the top causes of illness and death globally, particularly among children, the elderly, and people with immunocompromised conditions. It requires prompt and precise diagnosis for effective management. Chest X-rays and CT scans are standardized diagnostic modalities for pneumonia; however, they are expensive and take a lot of time. This creates the need for automated systems that utilize AI and DL technologies. The focus of most studies within this context relates to the application of CNNs and earlier versions of ViTs for CAD systems designed for pneumonia detection. While these methods have shown some level of effectiveness, they tend to overlook the long-range dependencies that exist in medical images due to their low computational efficiency, which limits their applicability in real-world healthcare scenarios. This is the problem that the current research aims to tackle by implementing The Swin-T-McSWA Transformer’s framework. It attempts to integrate the hierarchical design of the Swin Transformer with Multi-Context Shifted Window Attention (McSWA) and a lightweight T-Block to enhance performance further. To achieve these objectives, the study proposed the introduction of a new framework: Temporal Space-Driven Multi-Context Shifted Window Swin Transformer (Swin-T-McSWA). The McSWA mechanism improves the spatial feature extraction by contextually dependent window shifting, thereby capturing both local and global relationships in the medical image. The T-Block maintains accuracy critical for pneumonia feature detection, improves computational efficiency using dilated convolutions. The most notable aspect of this approach is how well it balances feature extraction with the associated computational expense. This produces a model with high accuracy and low expenditure of resources. As experiment results have shown, Swin-T-McSWA achieves outstanding accuracy of 98.76% and an F1 score of 98.17% in comparison with traditional CNNs and ViTs. More remarkably, the proposed model surpasses the EfficientNet-B0 model, which is often used for image classification, by 4.5% in accuracy (formerly 94.2% accuracy). In addition, the Swin-T-McSWA model achieved these results with just 37 M parameters and 7.2 GFLOPs, rendering the model computationally efficient and suitable for deployment in resource-limited environments. In conclusion, the Swin-T-McSWA model enhances automated systems for pneumonia detection, exceeding the accuracy and efficiency of existing models. The innovation has the potential to transform the field of medical image analysis while providing significant benefits in clinical contexts where rapid and accurate detection of pneumonia is essential for effective patient management.</p>