Scientific Session 05 — Neuroradiology - BrainMonday, May 1, 2017
2949. Diagnostic Accuracy of Paired Color and Gray-scale CT Images for Detecting Acute Ischemic Stroke
Swaminathan N*, Khan M, Lirette S, Griswold M, Smith A. University of Mississippi Medical Center, Jackson, MS
Address correspondence to A. Smith (firstname.lastname@example.org)
Objective: Unenhanced CT imaging is an essential component of the “Code Gray” protocol. Head CT images displayed in gray scale have only moderate diagnostic accuracy for depicting acute ischemic stroke. The purpose of this study was to improve diagnostic accuracy for detection of acute ischemic stroke in the setting of the Code Gray protocol by evaluating paired color and gray-scale head CT images.
Materials and Methods: For this institutional review board–approved, single-center, multireader retrospective study, we included adult patients who presented with acute strokelike symptoms triggering a Code Gray protocol who underwent head CT between January 1, 2010, and April 15, 2014, and a reference standard confirmatory MRI (n = 290). Using a test cohort (n = 25), custom, fully automated software was designed to color the intracranial contents on the head CT images and increase conspicuity of ischemic strokes. Using a separate validation cohort (n = 100), two imaging sets were generated from each patient’s head CT examination: gray-scale (containing only gray-scale images) and color and gray-scale (containing paired color and gray-scale images). A group of 14 readers (four neuroradiologists, four body radiologists, and six radiology residents) underwent 1 hour of training. Readers independently participated in two reading sessions and evaluated all patients twice, once with gray-scale images and once with paired color and grayscale images. Stroke findings were recorded in a web-based research electronic data capture database, and all reading session were timed. A cross-over design, patient randomization, and 4-week washout period were used to reduce recall bias between the reading sessions. The mean accuracy, sensitivity, and specificity for each reader group was compared between the gray-scale and color and gray-scale imaging sets. The dichotomous outcome on presence or absence of stroke was analyzed in a multivariate model.
Results: Among the 14 readers, the mean accuracy, sensitivity, and specificity for diagnosing acute ischemic stroke were 60%, 38%, and 92% using the gray-scale images and 63%, 46%, and 89% using the paired color and grayscale images, respectively (p < 0.001, p < 0.001, p = 0.018, respectively). The odds of correctly diagnosing any ischemic stroke (n = 60) and any large vessel ischemic stroke (n = 39) using paired color and grayscale versus gray-scale images increased by 39% and 51%, respectively (odds ratio [OR], 1.39 vs 1.51; 95% CI, 1.21–1.59 vs 1.29–1.77; p < 0.001 for both). The mean accuracy of residents, body imagers, and neuroradiologists was 58%, 60%, and 63%, respectively, using gray-scale images and increased in each group to 62%, 63%, and 66%, respectively, using paired color and gray-scale images (p = 0.019, 0.051, and 0.035, respectively). Mean interpretation time increased from 113 seconds with gray-scale images to 124 seconds using paired color and grayscale images (p < 0.001).
Conclusion: The use of paired color and gray-scale head CT images improved diagnostic accuracy for detection of acute ischemic stroke in the setting of a Code Gray protocol, and the accuracy increased at all reader experience levels. Given that 2 million head CT examinations are performed annually to evaluate strokelike symptoms in the United States alone, any increase in diagnostic accuracy would have wide implications.