ARRS 2022 Abstracts


1101. Dual-Energy and Oral Contrast: Improving Reader Accuracy and Confidence
Authors * Denotes Presenting Author
  1. Shaun Hinen; Medical University of South Carolina
  2. Madison Kocher *; Medical University of South Carolina
  3. Jordan Chamberlin; Medical University of South Carolina
  4. Jeffrey Waltz; Medical University of South Carolina
  5. Tristan Young; Medical University of South Carolina
  6. Veronica Krull; Medical University of South Carolina
  7. Andrew Hardie; Medical University of South Carolina
Computed tomography (CT) has generally become the preferred diagnostic imaging modality to assess for perforated viscous. Prior studies have demonstrated advantages in the use of CT performed with oral contrast over fluoroscopic equivalent examinations with generally higher sensitivity but lower specificity for CT over fluoroscopy. One potential advantage of performing CT over fluoroscopy that has not been explored extensively in the literature is the added value of dual-energy CT (DECT) acquisition. The purpose of this study was to assess the added value of using DECT iodine overlays (IO) as opposed to routine contrast-enhanced CT in the assessment of oral contrast leak.

Materials and Methods:
Three radiology residents were tasked with assessing DECT studies both with and without IO for the presence of a leak. Rater values were aggregated and weighted by years of experience (4, 2, and 1, respectively) and a total of 90 paired reads were then analyzed. Results were defined binarily as “leak” or “no leak.” Raters were also asked to assign subjective confidence scores to both contrast CT and IO studies which were assessed in random order with rater times measured in seconds. Unweighted aggregate diagnostic parameters were calculated with 95% confidence intervals given by the Clopper-Pearson method. Reliability analysis was performed with both unweighted and weighted Cohen’s Kappa values with 95% confidence intervals. Univariate analysis was performed using Mann-Whitney U tests and Wilcox signed-rank tests for continuous and ordinal variables, respectively. The Kruskal-Wallis test was used for comparison of multiple raters. Simple logistic regression was used along with DeLong’s test of correlated ROC curves for prediction of bowel leak. All a = 0.05. All analysis was performed in R v 3.6.3.

Reader 1 had an accuracy of 70% for routine CT and 83% for IO, whereas reader 2 had an accuracy of 73% for routine CT and 93% for IO. The most experienced reader, reader 3, had an accuracy of 100% for both. When pooled, accuracy of routine CT was 81.1% (95% CI 73.0 - 89.3%) for identifying a leak (sensitivity 0.84, specificity 0.78, PPV 0.79, and NPV 0.83) whereas IO was 92.2% accurate (95% CI 86.6-97.8%) (sensitivity 0.93, specificity 0.91, PPV 0.91, NPV 0.93). Aggregate weighted agreement between the readers revealed a kappa of 0.886 with IO, indicating excellent agreement. Iodine mapping was significantly better at predicting leak than contrasted CT (AUC = 0.922 vs. 0.811; p = 0.0103). All readers required significantly less time to interpret IO than routine CT regardless of the difference in experience. Reader 3 reported greater confidence in IO despite a perfect diagnostic performance, and reader 2 had an increased diagnostic confidence in IO while also realizing an improved performance.

Use of DECT iodine overlay is associated with increased accuracy for identification of oral contrast leak compared to contrast-enhanced CT with decreased time of interpretation and increased diagnostic confidence.