Akademik Veri Yönetim Sistemi
JOURNAL OF POLYTECHNIC-POLITEKNIK DERGISI, 2025 (ESCI, TRDizin)
The COVID-19 pandemic has underscored the urgent need for rapid and accurate diagnostic tools to support early intervention and containment. While chest X-ray (CXR) imaging has emerged as a practical modality for COVID-19 detection, existing studies often focus on binary classification or rely on single-model evaluations without addressing class imbalance, generalizability, or multi-class diagnostic scenarios. This study proposes a novel, standardized deep learning-based framework that classifies CXR images into three clinically relevant categories: COVID-19, Normal, and Pneumonia. Unlike previous works, our approach comprehensively evaluates and compares the performance of eight prominent deep learning models-CNN, ResNet50, Xception, DenseNet, MobileNet, VGG16, ResNet152v2, and InceptionV3-using a preprocessed dataset. Key innovations include the use of a unified preprocessing pipeline, class-balancing strategy, and detailed model comparison based on a rich set of evaluation metrics (Accuracy, Precision, Recall, F1-Score, FPR, FNR, and Specificity). The results demonstrate that MobileNet and VGG16 offer high diagnostic performance with low computational overhead, making them ideal for deployment in resource-limited clinical settings. Our study's uniqueness lies in its multi-model, multi-class evaluation design, interpretability through confusion matrix analysis, and robust benchmarking against real-world challenges such as class imbalance. This positions the proposed framework as a reliable and scalable CAD tool to aid frontline healthcare providers in the early detection and differential diagnosis of COVID-19 and other respiratory illnesses.