A Novel Approach for Classifying COVID-19 Based on LBP-DCNN Feature Fusion and Stacking Heterogeneous Ensemble Learning

Background and Objective: Accurate identification of suspected Coronavirus disease (COVID-19) cases is of great significance in controlling the spread of the disease and timely treatment. The main purpose of this study is to propose an effective diagnosis model to detect COVID-19.Methods: We propose an automated COVID-19 diagnosis process using Stacking-based integrated classification with a fusion-based feature extraction model, called LBP-DCNN. Firstly, the local binary pattern (LBP) algorithm is used to extract the detailed texture features of the CT image. The depth features are extracted using DenseNet201, a deep neural network pre-trained on the ImageNet dataset. Next, we use a parallel fusion method to fuse texture features and deep convolutional neural network (DCNN) features. Finally, propose a Stacking-based integrated classification method, selection of base learners and meta-learners through extensive experimentation. The performance of our proposed classification method is compared with other potential strategies and state-of-the-art models.Results: The proposed method was trained and tested on the prepared dataset, and the experimental results show that this approach achieves an overall accuracy of 99.27%, precision of 98.40% and sensitivity of 99.83%. In addition, the method outperformed 9 state-of-the-art COVID-19 detection methods.Conclusion: This study proposed a COVID-19 diagnosis model obtained promising results using first-line clinical imaging, and it can help radiologists to make accurate diagnoses based on CT images.