JMIR Publications

ABSTRACT

Background:

The aim of the present study is to build machine learning (ML) architectures, exploiting CT radiomics information, and analyze algorithms’ capability to predict over hospitalization stay of patients affected by Covid-19.

Objective:

The aim of the present study is to build machine learning architectures, exploiting CT radiomics information, that is able to predict hospitalization stay of patients affected by Covid-19, since the hospital admission.

Methods:

Original CT lung images of 38 Covid-19 patients underwent two segmentation phases in order to obtain ground glass area of the lung parenchyma. After an isotropic voxel resampling and wavelet and Laplacian of gaussian filtering, 92 intensity and texture radiomics features were extracted. Feature reduction was conducted by applying Last Absolute Shrinkage and Selection Operator (LASSO) to the radiomic features set. We trained three ML classification algorithms, such as Support Vector Machine (SVM), Quadratic Discriminant Analysis (QDA) and Random Forest (RF), and validated through 5-fold cross validation technique. Accuracy, sensitivity, specificity and area under the receiving operating characteristic were used to evaluate classification performance.

Results:

SVM classifier shows the highest average accuracy (92.1%), specificity (96.3%) and sensitivity (81.8%), using the original dimension feature set. Using the reduced dimension feature set (LASSO), RF was able to discriminate better over classes, reaching 94.7% in accuracy, 96.3% in specificity and 90.9% in sensitivity.

Conclusions:

The study shows that radiomic features can be used to build a machine learning framework for predicting patient hospitalization duration. Although some limitations occur, i.e. small size of the sample included, the study represents a relevant attempt for the application of CT radiomics for robust prognostic modeling to support health professionals and hospital management team to a better understand diseases and identification of effective treatment options. Clinical Trial: The Report-Age protocol study has been approved by the Ethics Committee of the IRCCS INRCA, Ancona, Italy (reference number CEINRCA-20008) and registered under the ClinicalTrials.gov database (reference number NCT04348396).