A Hybrid LBP and CNN-Based Approach for COVID-19 Detection Using Chest X-Ray Images

Given the global spread of COVID-19, there is an urgent need for accurate point-of-care tests in settings where RT-PCR has a long turnaround time or is unavailable. Early screening with chest X-ray (CXR) imaging is attractive because it is widely available and low cost. Nevertheless, the manual interpretation of X-ray images is time consuming, subjective and necessitates expert radiological skills. We propose a lightweight, efficient automated COVID-19 detection model by combining handcrafted features with deep learning features to classify CXR images. Preprocessing normalizes image dimensions and enhances contrast. Local Binary Pattern (LBP) features are employed to enrapture local texture information of lung regions. For context-aware representation, we extract high-level features with a pretrained CNN (ResNet-50). The LBP and CNN features are fused into a single feature vector, which is fed into a support vector machine (SVM) classifier. We evaluated our proposed model on a public dataset available for COVID-19 radiography, containing normal, pneumonia, and other confirmed COVID-19 cases. The results show high classification performance with an accuracy of 97.91%, precision of 97.8%, recall of 97.83% and F1-score of 97.23%. The large burden of clinical screening demonstrates that the fusion of LBP and CNN features dramatically promotes the differentiability between infectious patients with COVID-19 and non-infected cases, which provides a valid approach to address scalable accuracy in clinical examination.