Deep Learning Solutions for Pneumonia Detection: Performance Comparison of Custom and Transfer Learning Models

Pneumonia is one of the leading causes of illness and death worldwide. In clinical practice, Chest X-ray imaging is a common method used to diagnose pneumonia. However, traditional pneumonia diagnosis through X-ray analysis requires manual annotation by healthcare professionals which delays diagnosis and treatment. This study aimed to investigate and compare three different deep learning methodologies for classifying pneumonia to detect the disease in patients. These advanced models have the potential to overcome the challenges of reliability and accessibility of diagnostic practices. The methodologies evaluated included a custom convolutional neural network (CNN), a transfer learning approach using the ResNet152V2 architecture, and a fine-tuning strategy also based on ResNet152V2. The models were rigorously assessed and compared across various metrics, including testing accuracy, loss, precision, F1 score, and recall. The comparative analysis shows that the fine-tuning strategy outperforms the other methods in terms of operational effectiveness, with the custom CNN being the next most effective, and the transfer learning method ranking last. The study also highlights that false negatives can have more serious consequences than false positives, even without specialized medical knowledge.