Utility of image fusion software in identifying shunt malfunction.

Abstract

Pediatric shunt malfunction occurs frequently and is important to recognize due to the high associated morbidity and mortality. Although neuroimaging plays a crucial role in the diagnosis, it remains imperfect. We sought to identify the effect of image fusion software in predicting shunt malfunction. A total of 248 rapid shunt series brain MRIs performed between 2013 and 2017 were compared with prior neuroimaging for changes in ventricular size by two methods: radiology report and Brainlab fusion. Shunt malfunction was defined by an operative report confirming malfunction within 72h of neuroimaging. The two methods were compared by logistic regression models, with sensitivity and specificity subsequently calculated. Shunt malfunction was identified in 40 cases (16.1%). Imaging report demonstrated a lower Akaike information criterion than the Brainlab fusion and is therefore a better fitting model. While sensitivity is similar for the two models, 0.94 (0.90 to 0.97, 95% CI) for imaging report, and 0.95 (0.91 to 0.98, 95% CI) for Brainlab, the specificity was significantly different, 0.50 (0.37 to 0.63, 95% CI) and 0.33 (0.24 to 0.44, 95% CI) respectively. Our data indicate that an increased ability to detect subtle changes in ventricular size does not translate to improved accuracy, but instead leads to decreased specificity, and therefore an overdiagnosis of shunt malfunction in children with normally functioning shunts. While imaging continues to play a prominent role in the identification of shunt malfunction, neurosurgical clinical evaluation remains crucial to the final diagnosis.