Atıf İçin Kopyala
Dincer N., Görgel P.
Biomedical and Biotechnology Research Journal, cilt.9, sa.1, ss.24-29, 2025 (ESCI)
Özet
Background:
Coronavirus disease 2019 (COVID-19) is a respiratory disease seen in the lungs, while pneumonia is an inflammation seen in the lung tissue. The fact that the appearances of both diseases are similar in the medical images increases the importance of making their diagnosis correctly. In recent periods, the increase in deaths due to COVID-19 has led to an interest in studies related to early diagnosis of this disease. In addition to medical studies, computer-aided studies provide great support for early diagnosis.
Methods:
In this study, a model called Unsharp Swin transformer and Vision transformer network with Mobile Network Version 2 (MobileNetV2) (US-VM) was developed to classify the lung images. To test the proposed model, an original dataset was created by collecting images from different open-source data sets with COVID-19, normal, and pneumonia features. The proposed US-VM model was applied to the augmented version of this data set which was created by applying geometric transformations such as zooming, rotating, and cropping to the original images. Classical unsharp masking was added to the Swin transformer blocks as a part of the model and vision transformer was enhanced with MobileNetV2.
Results:
Successful classification results were obtained according to the performance evaluation of the proposed model via accuracy, F1-score, specificity, precision, and recall metrics.
Conclusions:
Our study demonstrates its success when compared to the studies with classical deep learning models in the literature. Furthermore, it is seen that the proposed system’s accuracy surpasses the model in which Swin and Vision transformers were used alone separately.