CT-Based Radiomic Analysis May Predict Bacteriological Features of Infected Intraperitoneal Fluid Collections after Gastric Cancer Surgery

Abstract

The ability of texture analysis (TA) features to discriminate between different types of infected fluid collections, as seen on computed tomography (CT) images, has never been investigated. The study comprised forty patients who had pathological post-operative fluid collections following gastric cancer surgery and underwent CT scans. Patients were separated into six groups based on advanced microbiological analysis of the fluid: mono bacterial (n = 16)/multiple-bacterial (n = 24)/fungal (n = 14)/non-fungal (n = 26) infection and drug susceptibility tests into: multiple drug-resistance bacteria (n = 23) and non-resistant bacteria (n = 17). Dedicated software was used to extract the collections’ TA parameters. The parameters obtained were used to compare fungal and non-fungal infections, mono-bacterial and multiple-bacterial infections, and multiresistant and non-resistant infections. Univariate and receiver operating characteristic analyses and the calculation of sensitivity (Se) and specificity (Sp) were used to identify the best-suited parameters for distinguishing between the selected groups. TA parameters were able to differentiate between fungal and non-fungal collections (ATeta3, p = 0.02; 55% Se, 100% Sp), mono and multiple-bacterial (CN2D6AngScMom, p = 0.03); 80% Se, 64.29% Sp) and between multiresistant and non-multiresistant collections (CN2D6Contrast, p = 0.04; 100% Se, 50% Sp). CT-based TA can statistically differentiate between different types of infected fluid collections. However, it is unclear which of the fluids’ micro or macroscopic features are reflected by the texture parameters. In addition, this cohort is used as a training cohort for the imaging algorithm, with further validation cohorts being required to confirm the changes detected by the algorithm.

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

“…21 Texture analysis of fluid collections has only been performed in few previous studies. 20,21,29 In a previous study on 320 patients, texture features were not diagnostic superior to conventional imaging findings to discriminate infected from sterile fluid collections. 20 However, in this study, bacterial and fungi were not further investigated by subanalysis, which could result in these negative results.…”

“…20,21,29 In a previous study on 320 patients, texture features were not diagnostic superior to conventional imaging findings to discriminate infected from sterile fluid collections. 20 However, in this study, bacterial and fungi were not further investigated by subanalysis, which could result in these negative results. Based upon this study, it was further deliberated that different microbiological pathogens might result in different CT texture features.…”

“…Finally, by the same work group, "ATeta3" and "ATeta4" were highlighted as possible texture features to discriminate bacterial from fungi infected fluid collections based upon 42 patients. 20 The three identified texture features of the present study were of different groups. The feature kurtosis is a first order histogram parameter, which reflects the slope of the histogram of the HUs.…”

“…To date, the exact diagnosis of underlying causes pathogens of an infected fluid collection is beyond the scope of imaging and can be provided by microbiological analysis only 20 . That is the reason why few systematic data exist to discriminate fungal and bacterial infected fluid collections based upon imaging 20 .…”

“…20 That is the reason why few systematic data exist to discriminate fungal and bacterial infected fluid collections based upon imaging. 20 In clinical routine, it could be of interest to predict the underlying cause, as treatment options differ significantly between fungal and non-fungal fluid collections. Moreover, it could enable a new method to use the radiological images.…”