Detection and classification of pneumonia in chest radiographs using hybrid CNN-Vision transformer model

ABSTRACT: Pneumonia remains one of the leading causes of mortality in the Philippines, with accurate and timely diagnosis hindered by the limited availability of radiologists and access to imaging technologies. This study developed a portable screening system that detects and classifies pneumonia in chest radiographs through a hybrid artificial intelligence model that combines Convolutional Neural Networks and Vision Transformer architectures. The device incorporates a microcomputer, high-resolution camera, and touchscreen interface to scan chest X-ray films, printed digital images, or soft copies, and it provides real-time classifications such as normal, bacterial pneumonia, viral pneumonia, and other pathological findings. A PDF report is automatically generated and sent to radiologists for expert confirmation. The dataset used consisted of over 9000 images and was validated by two radiologists, Dr. Justin Luke Yap and Dr. Michael Angelo A. Gemarino. The hybrid model achieved a model accuracy of 80 percent, which was the lowest among the related studies considered. This difference is attributed to the inclusion of more diverse inputs such as scanned films and printed copies, whereas other studies relied solely on online datasets. Despite this, the system achieved a higher accuracy of 93 percent, reflecting strong performance in real-world conditions. Healthcare professionals evaluated the system as reliable, portable, and easy to use, citing user-friendly graphical interface and image enhancements capabilities. These results indicate that the device is a viable support tool for initial pneumonia screening, especially in underserved communities where radiological expertise and resources are limited.

Filipiniana