Abstract

Pneumonia is one of the lung infections that has a significant impact on global health due to its high mortality rate, thus requiring a fast and accurate early diagnosis system. This study aims to develop an automatic pneumonia detection system using a Convolutional Neural Network (CNN) architecture tailored for chest X-ray image classification. The model consists of three sequential convolutional blocks with 32, 64, and 128 filters, respectively, and employs a Global Max Pooling layer. The dataset used comprises 5,856 X-ray images divided into pneumonia and normal classes. Experiments were conducted by comparing two optimization algorithms: RMSProp and SGD. Training results showed that SGD achieved the best performance, with validation accuracy reaching $93.19 \%$, outperforming RMSProp, which only achieved a lower accuracy rate. These findings indicate that the combination of a specialized CNN architecture and the appropriate use of the SGD optimizer can produce an accurate and efficient pneumonia classification system with the potential to support automated clinical diagnosis processes.

Concepts :

Access to Document

Citations by Year