The article discusses the challenges of using deep learning methods to detect COVID-19 using public datasets of chest X-rays and CT scans. The researchers found that many studies show high accuracy in classifying COVID-19 patients, but these results may be biased and overestimated due to the way the experiments are designed. By combining two popular classification networks, the researchers were able to identify biases and overfitting issues in the deep learning models. They suggest that a larger, less biased database is needed for developing tools that can be used in real clinical settings.