ML project for Covid-19
We are currently working on multiple ML project for managing the COVID-19 crisis.
We developed EMR based fusion AI models to predict if a COVID-19 patient will need hospitalization within 7 days of the RT-PCR test. Compare to the existing Covid-19 prediction models, our work has a much broader scope and many of the critical event prediction models rely on very recently generated medical information of the patients and predict the event for a very tight time-window in time for future, varying between 24 hours to 7 days. Hence, patients must already be utilizing healthcare resources in a healthcare facility when these models can be applied for prediction. Instead, our model relies on past health records of patients, shunning any information generated during their current COVID-19-related hospital visit. This enables our model to be applied before COVID-19 infected patient uses any healthcare resources, making it an effective future resource utilization predictor. Early fusion is the best performing model with an 84 overall f1-score [CI 82.1-86.1] and 85 f1-score for identifying patients who will need hospitalization within 7 days of RT-PCR testing.
We are also validating the efficiency and applicability of existing deep learning vision models for COVID-19 diagnosis from Chest X-ray. We provide results evaluated on three publicly available datasets (created before the COVID-19 outbreak) and an additional institutional dataset collected from Emory University Hospital between January and Nov 2020, containing patients tested for COVID-19 infection using RT-PCR.