An analysis of radiomics features in lung lesions in COVID-19

Abstract

Radiomic features extracted from CT imaging can be used to quantitively assess COVID-19. The objective of this work was to extract and analyze radiomics features in RT-PRC confirmed COVID-19 cases to identify relevant characteristics for COVID-19 diagnosis, prognosis, and treatment. We measured 29 morphology and second-order statistical-based radiomics features from 310 lung lesions extracted from 48 chest CT cases. Features were evaluated according to their coefficient of variation (CV). We calculated the CV for each feature under two statistical conditions: one with all lesions weighted equally and one with all cases weighted equally. In analyzing the patient data, there were 6.46 lesions-per-case and for 81.25% of cases, the lesions presented with bilateral lung involvement. For all radiomic features examined except ‘energy’, the CV was higher in the lesion distribution than the case distribution. The CV for morphological features were larger than second-order in both distributions, 181% and 85% versus 50% and 42%, respectively. The most variable features were ‘surface area’, ‘ellipsoid volume’, ‘ellipsoid surface area’, ‘volume’, and ‘approximate volume’, which deviated from the mean 173-255% in the lesion distribution and 119-176% in the case distribution. The features with the lowest CV were ‘homogeneity’, ‘discrete compactness’, ‘texture entropy’, ‘sum average’, and ‘elongation’, which deviated less than 31% by case and less than 25% by lesion. Future work will investigate integrating this data with similar studies and other diagnostic and prognostic criterion enhancing the role of CT in detecting and managing COVID-19.