Computed tomography texture analysis to discriminate fungal and non-fungal infected fluid collections

Abstract

Purpose: Texture analysis of computed tomography (CT) can aid in characterization of fluid collections providing biomarkers. The present study tested whether texture analysis can discriminate between fungal or non-fungal infection in patients undergoing CT-guided percutaneous drainage treatment.

Approach: Overall, 214 patients [($n=76$ females, 35.5%); mean age $62\pm 14$ years and range 20 to 94 years] with 255 fluid collections were included in the analysis. All patients underwent CT-guided drainage treatment and were evaluated with microbiological analysis. CT texture analysis was performed with the MaZda package.

Results: Only three of the investigated CT texture features were statistically significant different between the groups, namely kurtosis ($p=0.04$), S(3,3)InvDfMom ($p=0.02$), and S(5,-5)DifEntrp ($p=0.003$). These texture features were further investigated by the receiver operating characteristic curve. S(3,3)InvDfMom achieved the highest accuracy with an area under the curve of 0.62, resulting in a sensitivity of 0.66 and a specificity of 0.57.

Conclusion: Some CT texture features were different between fungal and non-fungal infected fluid collections. The diagnostic overlap is large, which could reduce the clinical benefit. Further studies are needed to identify the possible diagnostic benefit of texture analysis in these patients.

Keywords: computed tomography; fluid collection; fungal infection; texture analysis.