Deep learning-based fine-grained automated pneumonia detection model

• Main Authors: Rangasamy, Keerthana, Mohd. Fuzi, Nurul Shuhada, Amir As’Ari, Muhammad, Rahmad, Nur Azmina, Sufri, Nur Anis Jasmin
• Format: Article
• Language: English
• Published: Taylor's University 2022
• Subjects: Q Science (General)
• Online Access: http://eprints.utm.my/103130/1/KeerthanaRangasamy2022_DeepLearning%20BasedFine%20Grained.pdf; http://eprints.utm.my/103130/; https://jestec.taylors.edu.my/Vol%2017%20Issue%204%20August%202022/17_4_09.pdf
• Institution: Universiti Teknologi Malaysia

Pneumonia is a bacterial, fungal, or viral infection of the lungs that leads the lungs' air sacs to clogged with pus or fluids that are generally diagnosed using chest X-rays (CXR) cost-effective, fast, and non-invasive. However, this diagnosis is complicated by high inter-observer and intra-observer variation among radiologists as it mainly depends on radiologist proficiency. Hence, there is a higher demand for automated, rapid pneumonia detection tools to curb the lack of specialised radiologists, especially in rural areas. Thus, this paper presented a fine-grained deep learning-based automated pneumonia detection system using several well-establish pre-trained Convolutional Neural Network (CNN) models (AlexNet, SqueezeNet, GoogleNet, ResNet-18, and ResNet-50) form CXR images that can be utilised for early diagnosis. The results revealed that all models succeed in detecting pneumonia at an accuracy of over 80%. SquuezeNet outperformed among the other models with an accuracy of 81.62% within a speed of 64.6 minutes.