IMI/Publicaţii/CSJM/Ediţii/CSJM v.30, n.2 (89), 2022/

A Comparative study on classification performance of Emphysema with transfer learning methods in deep convolutional neural networks

Authors: Selçuk Yazar
Keywords: Deep Learning, Transfer Learning, Convolutional Neural Networks, Medical Imaging, Emphysema Diagnosis.


Today Emphysema, which takes place among the top five diseases, is encountered in the western world in terms of rehabilitation and healthcare costs. Diagnosis of this type of respiratory tract disease with the help of computers is gradually increasing its importance. In this study, we aimed to classify it with the transfer learning method by using single labeled emphysema diagnosed data which is obtained from three large data sets. We classified the images that are obtained from ChestX-ray14, CheXpert, and PadChest databases by 95\% of Area Under the Curve (AUC) with the fully connected layer model and DenseNet-121 pre-trained neural network and 90\% of Area Under the Curve (AUC) with Xception pre-trained neural network. We evaluated this proposed deep learning-based model as an effective and practical diagnostic tool for emphysema alone, using x-ray data. Notably, transfer learning is a very functional approach in terms of differentiation between normal and patient in similar diseases that have just emerged in the pandemic period that we live in.

Kırklareli University Software Engineering Department
Kırklareli / Turkey



Adobe PDF document0.28 Mb