DEVELOPMENT OF A PREDICTIVE MODEL FOR THE DETECTION OF PNEUMONIA USING TRANSFER LEARNING.
Abstract
This study aims to identify the most efficient transfer learning strategies for improving the diagnostic performance in medical imaging applications of pictures obtained from chest x-rays. A dataset containing chest X-ray (CXR) images of both healthy individuals and those with pneumonia was identified and collected from an online public repository following which the image dataset was subjected to data preparation. The study adopted the use of an ensemble of four transfer learning techniques, namely, VGG16, CGG19, ResNet50, and GoogleNet network architectures based on a voting approach. The results of the study revealed that the pre-trained models that were adopted for this study were able to perform well in the detection of normal and pneumonia CXR images. This study also revealed that increasing the size of the train dataset that was used for model building caused an improvement in the performance of the pretrained models during evaluation using the testing dataset.
Full-Text Access Notice
In accordance with the NERD Policy on promoting peer-reviewed publication, public access to the full text of a project, thesis or dissertation is restricted for three years, allowing the author and supervisors sufficient time to pursue peer-reviewed publication.
During this period, researchers with legitimate academic or research purposes may request authorisation directly from the author to enable NERD to release the indexed full texts of the work using the form below.