Comparison of deep learning architectures for CXR opacity detection

• Main Authors: Rafiqin Roslan, Siti Nurhajar, Che Azemin, Mohd Zulfaezal, Md. Ali, Mohd. Adli, Mohd Tamrin, Mohd Izzuddin, Jamaludin, Iqbal
• Format: Conference or Workshop Item
• Language: English; English
• Published: Association for Computing Machinery (ACM) 2022
• Subjects: RC731 Specialties of Internal Medicine-Diseases of The Respiratory System; TK7885 Computer engineering
• Online Access: http://irep.iium.edu.my/98251/7/98251_Comparison%20of%20Deep%20Learning%20Architectures%20for%20CXR%20Opacity%20Detection.pdf; http://irep.iium.edu.my/98251/8/98251_Screenshot%20of%20the%20Proceedings.pdf; http://irep.iium.edu.my/98251/; https://dl.acm.org/doi/10.1145/3524304.3524316; https://doi.org/10.1145/3524304.3524316

Previous research has shown that x-ray images can be labeled based on their abnormalities. The problem with the labels includes inconsistencies in the assignment of the abnormality which may lead to overestimation of the model performance. To overcome the problem of the majority-vote approach, adjudicated labels could be used. This researchwork highlights the comparison of deep learning architectures for chest x-ray opacity detection. This study aims to investigate the best performance of the different deep learning models used when they are trained with the publicly available deep learning architectures and data set rather than using one type of deep learning model. Among the different deep learning architectures models used, an optimal model would be identified based on the best performance metrics.