Publication Date: 2020/02/19
Abstract: In diagnosing pneumonia, a physician needs to perform a series of tests, one of which is by manually examining a patient's chest radiograph. In the case of a large amount of data, errors in the diagnostic process could occur due to human error and this, of course, can endanger the patient's life. Moreover, the conventional method above is also quite time-consuming. In this study, research was conducted to train classifiers using Convolutional Neural Network (CNN) to automatically recognize normal chest radiographs and chest radiographs with pneumonia. Several architectures are used to train the classifier from previous papers references that already proven to have high accuracy, namely VGG16, InceptionV3, VGG19, DenseNet121, Xception, and ResNet50. Besides that, we added data augmentation to this training. As the results, VGG16 architecture has the highest accuracy with training accuracy reaching 0.9824% and validation accuracy 0.9215% therefore, VGG16 could be the best option among the other architectures in automatically recognizing pneumonia from chest radiograph images.
Keywords: Convolutional Neural Network, Deep Learning, Image Classification, Pneumonia Detection.
DOI: No DOI Available
PDF: https://ijirst.demo4.arinfotech.co/assets/upload/files/IJISRT20FEB134.pdf
REFERENCES