Layers Modification of Convolutional Neural Network for Pneumonia Detection
Wahyudi Setiawan, Fitri Damayanti
Abstract
Abstract Pneumonia is a bacterial, virus and fungi infection that attacks respiratory function. The disease causes air sacs in the lungs inflamed and swollen. It conditions produce lungs filled with fluid and mucus. Generally, the detection of pneumonia was done by chest x-ray images. This study discusses the detection of pneumonia through x-ray images using Convolutional Neural Network. The CNN model was Visual Group Geometry VGG16 and VGG19. As a comparison, we used the modified CNN 35 layer. The experiment using public data from Chest X-Ray Images - Kaggle. Data consist of 2 classes: normal and pneumonia with a total of 624 images. The results using VGG16 show a performance measure of sensitivity 92.75%, specificity 96.8%, and accuracy 94.1%. The result of VGG19 has sensitivity 96.6%, specificity 94.3%, and accuracy 95.7%. For CNN 35 layer has sensitivity 95.1%, specificity 98.5%, and accuracy 96.3%.