Deep Learning-Based Evaluation of ICU Requirements in COVID-19 Cases

According to WHO statistics, new COVID-19 virus has spread quickly worldwide since the first instance of infection in China, infecting a total of 442,602,593 people. In order to stop the virus from spreading, diagnosis is essential. Yet, the existing diagnostic options for COVID-19 are inadequate in light of the mounting infection rates. A model which is more generalizable and can detect such diseases with the use of fewer resources must also be developed because of the inadequacy and dearth of generalizable training diagnostic systems on big data as well as the inadequacy regarding effective and quick systems in mobile devices. Finding a quick method which could handle a lot of data with high accuracy and few false negatives or false positives is thus important for stopping the virus from spreading. In the presented work, a new mathematical model has been presented for introducing hybrid as well as deep learning-based machine learning (ML). It extracted features from chest X-ray (CXR) images using pre-trained VGG16 (First Method) and MobileNetV2 (Second Method) models, and then utilized linear discrimination analysis as a dimensionality reduction method for speeding up the classification process and an XGBOOST classifier to finish the task. The F1-score, accuracy, precision, and recall of the first approaches were (0.92), (0.94%), (0.93), and (0.93), respectively. The F1-score, accuracy, precision, and recall for the second technique were (0.95), (0.95), (0.95), and (0.96), respectively.