Iteratively Pruned Deep Learning Ensembles for COVID-19 Detection in Chest X-rays

摘要: We demonstrate use of iteratively pruned deep learning model ensembles for detecting pulmonary manifestation COVID-19 with chest X-rays. This disease is caused by the novel Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) virus, also known as (2019-nCoV). A custom convolutional neural network and a selection ImageNet pretrained models are trained evaluated at patient-level on publicly available CXR collections to learn modality-specific feature representations. The learned knowledge transferred fine-tuned improve performance generalization in related task classifying CXRs normal, showing bacterial pneumonia, or COVID-19-viral abnormalities. best performing reduce complexity memory efficiency. predictions best-performing combined through different ensemble strategies classification performance. Empirical evaluations that weighted average significantly improves resulting an accuracy 99.01% area under curve 0.9972 findings CXRs. transfer, iterative pruning, resulted improved predictions. expect this can be quickly adopted screening using radiographs.