<![CDATA[Background: Pneumonia remains one of the leading causes of morbidity and mortality worldwide, particularly among children and older adults in low-resource settings. Diagnosis based on chest X-ray interpretation often depends on radiologist expertise, which may be limited in availability and prone to subjectivity. Deep learning offers a promising alternative to improve diagnostic efficiency and consistency.Aims: This study aims to evaluate the effectiveness of the ResNet-50 architecture for pneumonia detection using chest X-ray images by comparing transfer learning with frozen layers and partial fine-tuning strategies.Methods: A total of 5,856 chest X-ray images were obtained from a public dataset and divided into training, validation, and testing sets using stratified sampling. Data preprocessing included resizing, normalization, and augmentation. Two models were developed: (1) a frozen ResNet-50 model, where all convolutional layers were fixed, and (2) a fine-tuned ResNet-50 model, where the final convolutional layers were retrained. Performance was evaluated using accuracy, precision, recall, F1-score, and area under the ROC curve (AUC). Statistical tests were conducted to assess performance differences between the two models.Results: The frozen model achieved an accuracy of 62.50% and an AUC of 0.4819, indicating weak classification performance. In contrast, the fine-tuned model demonstrated substantially higher accuracy of 85.90%, F1-score of 0.8967, and AUC of 0.9510, showing strong discriminative capability. Statistical analysis confirmed that the performance improvement in accuracy was significant.Conclusion: Fine-tuning significantly enhances the applicability of ResNet-50 for pneumonia detection. Without feature adaptation, pretrained models struggle to generalize to medical imaging domains. Fine-tuned transfer learning provides a more reliable framework for developing computer-aided diagnostic systems, particularly in clinical environments with limited expert availability.]]>
Copyrights © 2025
