120. 3D Ventricular Volume Analysis to Detect Shunt Failure
Features NREF-funded Author
Authors: Siri Sahib Singh Khalsa, MD; Jamaal Tarpeh, BS; David Altshuler, MD; Cormac Maher, MD (Ann Arbor, MI)
Introduction: The pre-operative diagnosis of shunt failure typically depends on detecting a difference in ventricular caliber between the current cranial imaging and the most recent well-scan. Patients with hydrocephalus often present with concerns for shunt failure to their local emergency department, which may not have access to previous cranial imaging for comparison. Radiology reports typically describe ventricular caliber in qualitative terms. Thus, a comparison of ventricular caliber by telephone is usually unreliable. We sought to develop and validate a computer program to rigorously measure the 3D ventricular volume on MRI and CT scans to make objective comparisons possible. Methods: A computer program was developed in Matlab to semi-automatically calculate the 3D volume of cerebrospinal fluid within the entire ventricular system on brain MRI and head CTs, within seconds. A combination of k-means clustering and nearest-neighbor techniques were used for the 3D segmentation. The algorithm was tested on 52 scans from 16 pediatric patients with shunted hydrocephalus. Shunt failure scans confirmed with subsequent shunt exploration were compared to the most recent well-scan. Well-scans were compared to the preceding well-scan for control. Results: Mean ventricular volume change was +600% +/- 212% for failure-scans and -2.5% +/- 9% for well-scans (p < 0.01). A ventricular volume increase cut-off of 20% was 100% sensitive and 89% specific for shunt failure in this cohort. Conclusion: A computer program was developed to rapidly calculate the volume of the ventricular system on both brain MRIs and head CTs. This method could be used to objectively detect shunt failure between institutions that are not able to share images. This benefit would apply only to patients whose ventricles enlarge during shunt failure. Prospective analysis of a large multi-center cohort will be necessary prior to implementation.