TY - GEN
T1 - Pneumonia Identification from Chest X-rays (CXR) Using Ensemble Deep Learning Approach
AU - Mun, Ng Weng
AU - Solihin, Mahmud Iwan
AU - Chow, Li Sze
AU - Machmudah, Affiani
N1 - Publisher Copyright:
© 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.
PY - 2022
Y1 - 2022
N2 - Chest x-ray screening has proven to be the most reliable method to diagnose pneumonia. However, it requires a professional radiologist to identify the symptom of pneumonia from each x-ray images. In the scarcity of professional radiologists, computer vision can assist in diagnosing x-ray images. This study aims to design a reliable image classifier for diagnosing pneumonia using an ensemble deep learning approach. Multiple experiments are conducted to evaluate transfer learning applications, data augmentations, and ensemble techniques. The pre-trained deep learning models are Xception, DenseNet201, ResNet152V2, InceptionResNetV2, NASNetLarge, and VGG16. The dataset used for training the models is obtained from Guangzhou Women and Children’s Medical centre. Each of the chosen models is trained and fine-tuned with Nesterov Stochastic Gradient Descent optimizer with their respective learning rate. The majority voting ensemble approach is employed to archive an accuracy of 97.56 and 99.14% for train and test data, respectively. It yields an F1 score of 99.25% for the test data.
AB - Chest x-ray screening has proven to be the most reliable method to diagnose pneumonia. However, it requires a professional radiologist to identify the symptom of pneumonia from each x-ray images. In the scarcity of professional radiologists, computer vision can assist in diagnosing x-ray images. This study aims to design a reliable image classifier for diagnosing pneumonia using an ensemble deep learning approach. Multiple experiments are conducted to evaluate transfer learning applications, data augmentations, and ensemble techniques. The pre-trained deep learning models are Xception, DenseNet201, ResNet152V2, InceptionResNetV2, NASNetLarge, and VGG16. The dataset used for training the models is obtained from Guangzhou Women and Children’s Medical centre. Each of the chosen models is trained and fine-tuned with Nesterov Stochastic Gradient Descent optimizer with their respective learning rate. The majority voting ensemble approach is employed to archive an accuracy of 97.56 and 99.14% for train and test data, respectively. It yields an F1 score of 99.25% for the test data.
KW - Chest x-ray image classification
KW - Deep learning
KW - Pneumonia detection
KW - Transfer learning
UR - http://www.scopus.com/inward/record.url?scp=85126958168&partnerID=8YFLogxK
U2 - 10.1007/978-981-16-8690-0_99
DO - 10.1007/978-981-16-8690-0_99
M3 - Conference contribution
AN - SCOPUS:85126958168
SN - 9789811686894
T3 - Lecture Notes in Electrical Engineering
SP - 1139
EP - 1151
BT - Proceedings of the 6th International Conference on Electrical, Control and Computer Engineering, InECCE 2021
A2 - Md. Zain, Zainah
A2 - Sulaiman, Mohd. Herwan
A2 - Mohamed, Amir Izzani
A2 - Bakar, Mohd. Shafie
A2 - Ramli, Mohd. Syakirin
PB - Springer Science and Business Media Deutschland GmbH
T2 - 6th International Conference on Electrical, Control and Computer Engineering, InECCE 2021
Y2 - 23 August 2021 through 23 August 2021
ER -