Abstract
<jats:p>The leading cause of mortality in children, it is crucial that clinicians immediately arrive at an accurate diagnosis. The suggested deep learning algorithm can automatically detect pneumonia using chest X-rays with a 97.7% accuracy rate. EfficientNetB4 provides an optimal trade-off between fast processing and diagnostic precision. The model's sensitivity to identify small lung abnormalities is enhanced by using Architectural Convolutional Layers. These layers extract morphological and connecting data that is not captured by ordinary convolutional layers. The approach improves model accuracy using MixUp, CutMix, and adaptive distortion approaches to handle tiny imbalanced pediatric datasets. To verify results, radiologists employ Grad-CAM graphics to examine model predictions, which highlight clinically significant areas. Sensitivity study shows that the model remains stable when presented with noisy or imbalanced data, indicating that it is ready for use in real-world clinical settings.</jats:p>