Pneumonia remains a leading global cause of morbidity and mortality, necessitating early and accurate diagnosis for effective treatment and improved patient outcomes. Deep learning (DL) models have shown remarkable potential in automating pneumonia detection from chest X-rays (CXR) and computed tomography (CT) scans, reducing radiologists’ workload and expediting the diagnostic process. This review evaluates the efficacy, sensitivity, and specificity of DL models for pneumonia detection by systematically analyzing 15 high-quality studies. Our findings indicate that many DL models achieve diagnostic accuracies exceeding 90%, with convolutional neural networks (CNNs) and transfer learning-based architectures demonstrating superior performance. These models promise deployment in clinical settings, particularly in resource-constrained environments with scarce radiological expertise. Despite these advancements, key challenges persist. Limitations include dataset biases, limited generalizability to diverse patient populations, and difficulties in real-world clinical integration. Many models are trained on high-quality, curated datasets that may not accurately reflect real-world imaging conditions, potentially affecting their reliability in routine clinical practice. Additionally, variability in study methodologies and a lack of standardized reporting metrics hinder direct comparison across models. Future research should address these limitations by developing more diverse and representative datasets, external validation across multiple healthcare settings, and enhanced interpretability of AI-driven diagnostic models. To facilitate widespread adoption, further studies are needed to explore model explainability, regulatory compliance, and integration strategies within clinical workflows. While DL models hold significant promise for pneumonia detection, their full clinical potential can only be realized through continued research, validation, and systematic implementation.