Abstract
Tuberculosis (TB) remains a life-threatening disease and is one of the leading causes of mortality in developing regions due to poverty and inadequate medical resources. Tuberculosis is medicable, but it necessitates early diagnosis through reliable screening techniques. Chest X-ray is a recommended screening procedure for identifying pulmonary abnormalities. Still, this recommendation is not enough without experienced radiologists to interpret the screening results, which forms part of the problems in rural communities. Consequently, various computer-aided diagnostic systems have been developed for the automatic detection of tuberculosis. However, their sensitivity and accuracy are still significant challenges that require constant improvement due to the severity of the disease. Hence, this study explores the application of a leading state-of-the-art convolutional neural network (EfficientNets) model for the classification of tuberculosis. Precisely, five variants of EfficientNets were fine-tuned and implemented on two prominent and publicly available chest X-ray datasets (Montgomery and Shenzhen). The experiments performed show that EfficientNet-B4 achieved the best accuracy of 92.33% and 94.35% on both datasets. These results were then improved through Ensemble learning and reached 97.44%. The performance recorded in this study portrays the efficiency of fine-tuning EfficientNets on medical imaging classification through Ensemble.
Highlights
According to the World Health Organization (WHO), tuberculosis (TB) is one of the leading causes of death [1, 2]
Deep convolutional neural networks (CNNs) models proposed in [27] were integrated with the handcraft technique through Ensemble learning as feature extractors from chest X-ray (CXR). e features extracted were used as inputs to train a classifier for detecting infected CXRs. e model was evaluated to compare the performance of both methods, and the Ensemble model performed better at 0.99 AUC
Each dataset is split into a 75% training set and 25% test set, respectively. e experiments were entirely performed using the Keras deep learning framework running on the TensorFlow backend
Summary
According to the World Health Organization (WHO), tuberculosis (TB) is one of the leading causes of death [1, 2]. Three pretrained CNN models are trained individually with different hyperparameters to classify the CXR into normal and abnormal classes. E study in [23] proposed an Ensemble of pretrained CNN models toward learning modality-specific features from different CXR collections. Deep CNN models proposed in [27] were integrated with the handcraft technique through Ensemble learning as feature extractors from CXR. E features extracted were used as inputs to train a classifier for detecting infected CXRs. e model was evaluated to compare the performance of both methods, and the Ensemble model performed better at 0.99 AUC. (1) Implementation of state-of-the-art EfficientNets to develop an effective and inexpensive TB detection system It is the first time the EfficientNet model is being ensembled to classify CXR images for tuberculosis diagnosis. E rest of the work is structured as follows: Section 2 presents detailed data and methodology explored in this study. e experimental results and discussion are provided in Section 3, while Section 4 concludes and gives insight into future direction
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
