Use of CNN with Transfer Learning to Improve the Accuracy of Pneumonia Diagnosis

Iparraguirre-Villanueva O.; Robles-Espiritu W.; Suxe-Ramírez M.; Flores-Castañeda R.O. · International Journal of Engineering Trends and Technology (2025) · DOI: 10.14445/22315381/ijett-v73i6p131

Resumen

Pneumonia is a respiratory disease affecting both lungs, causing symptoms such as cough with phlegm or pus, fever, chills and shortness of breath. Different microorganisms, such as bacteria, viruses and fungi, can cause pneumonia. Today, combating pneumonia is challenging for physicians, especially in vulnerable communities and during cold or sudden weather changes. In this work, we explored how Artificial Intelligence can contribute to improving the diagnosis of pneumonia. For this purpose, two Convolutional Neural Network (CNN) models, VGG16 and ResNet50-v2, were evaluated using the transfer learning technique to identify between healthy lungs and those affected by the disease. We worked with a dataset extracted from the Kaggle platform, which included 5216 images for training, 624 for testing and 16 for validation. The results showed that the ResNet50-v2 model obtained better results, reaching an accuracy rate of 90.87% in the test set, standing out for its ability to identify pneumonia cases. These results reinforce the potential of artificial intelligence as a diagnostic support tool. Finally, it can be stated that this work represents a significant advance in using neural networks in medicine. Integrating this technology into public health may not only improve the identification of pneumonia but may also contribute to the development of new efficient healthcare systems.

Instituciones: Universidad Tecnológica del Perú; Universidad César Vallejo