Machine and Deep Learning Algorithms for COVID-19 Mortality Prediction Using Clinical and Radiomic Features

Aim: Machine learning (ML) and deep (DL) predictive models have been employed widely in clinical settings. Their potential support aid to the clinician of providing an objective measure that can be shared among different centers enables possibility building more robust multicentric studies. This study aimed propose a user-friendly low-cost tool for COVID-19 mortality prediction using both ML DL approach. Method: We enrolled 2348 patients from several hospitals Province Reggio Emilia. Overall, 19 features were provided by Radiology Units Azienda USL-IRCCS Emilia, 5892 radiomic extracted each patient’s high-resolution computed tomography. built trained two classifiers predict mortality: machine algorithm, or vector (SVM), model, feedforward neural network (FNN). In order evaluate impact feature sets on final performance classifiers, we repeated training session three times, first only features, then employing finally combining information. Results: obtained similar performances algorithms, with best area under receiver operating characteristic (ROC) curve, AUC, exploiting information: 0.803 model 0.864 model. Conclusions: Our work, performed large heterogeneous datasets (i.e., data CT scanners), confirms results recent literature. Such algorithms included practice framework since they not applied but also other classification problems such as diabetic prediction, asthma cancer metastases prediction. proves lesion’s inhomogeneity depicted combined information is relevant