This paper explores the effectiveness of a hybrid CNN-Transformer model for pneumonia identification from chest X-ray pictures is investigated in this work. We obtain higher accuracy and enhanced localisation of regions afflicted by pneumonia by utilising Transformers’ global context awareness and CNNs’ spatial feature extraction capabilities. Furthermore, the hybrid loss function guarantees improved infection zone segmentation by fusing IoU and binary cross-entropy. Clinical investigations and current research show that our method works far better than traditional CNN models, leading to considerable gains in diagnosis accuracy. The results validate possible uses in real-time medical imaging systems as proposed in recent research.