Tehran University of Medical Sciences

Science Communicator Platform

Stay connected! Follow us on X network (Twitter):
Share By
A Modified Deep Learning Model in the Classification of Post-Covid-19 Lung Disease and a Comparative Study on Iranian and International Databases Publisher



Cheraghi S1, 2 ; Amiri S3 ; Abdolali F4 ; Esfahani AJ5, 6 ; Zade AAT7, 8 ; Ahadi R9 ; Ansari F2 ; Nafchi ER2 ; Hormozimoghaddam Z1, 2
Authors

Source: International Journal of Radiation Research Published:2024


Abstract

Background: We introduced Mask R-CNN+CNN as a deep learning model to classify COVID-19 and non-COVID-19 cases. Radiomic features relevant to COVID-19 was presented for Iranian and other nationalities. Materials and Methods: Chest CT images from 800 COVID-19 positive and negative patients were studied. The automated volume of the lung and segmentation of COVID-19 lung lesions were implemented using 3D U-net, Capsule network, and Mask R-CNN on annotated CT images. Deep learning models designed were based on Mask R-CNN, CNN, and Mask R -CNN+CNN algorithms to classify COVID-19 cases. We also explored radiomic features relevant to the COVID-19 pandemic in the lungs for chest CT images and implemented random forest (RF), decision tree (DT), and gradient boosting decision tree (GBDT) algorithms on two datasets. Results: The Mask R-CNN+CNN model demonstrated a higher classification accuracy (96.39 ± 2.94) compared to the Mask R-CNN and CNN models. The RF algorithm had greater power in differentiating relevant COVID-19 radiomic features compared to DT and GBDT, with an accuracy of at least 91 and an AUC of at least 985 in both datasets. We identified six radiomic features that were relevant to the pathological characteristics of COVID-19 positive/negative patients and were common across all datasets. Conclusion: This study emphasizes the power of Mask R-CNN+CNN with a ResNet-101 backbone as a CNN algorithm that utilizes bounding box offsets output from Mask R-CNN as the input for classifying COVID-19 cases. Radiomic features extracted from lung CT images might aid the diagnosis of COVID-19 in patients at various stages of the disease. © 2024 Novin Medical Radiation Institute. All rights reserved.
Other Related Docs
8. Segmentation of Covid-19 Pneumonia Lesions: A Deep Learning Approach, Medical Journal of the Islamic Republic of Iran (2020)
15. Application of Explainable Convolutional Neural Networks on the Differential Diagnosis of Covid-19 and Pneumonia Using Chest Radiograph, Proceedings of 2023 6th International Conference on Pattern Recognition and Image Analysis# IPRIA 2023 (2023)
17. The Role of Medical Imaging in Covid-19, Advances in Experimental Medicine and Biology (2021)
18. Prediction of Covid-19 Patients’ Survival by Deep Learning Approaches, Medical Journal of the Islamic Republic of Iran (2022)