MR QUANTIFICATION OF CEREBRAL VENTRICULAR VOLUME USING A SEMIAUTOMATED ALGORITHM

PURPOSE: A semiautomated border identification algorithm, insensitive to user bias, is evaluated for accuracy and speed in the measurement o f ventricular volumes from three-dimensional MR images. METHODS: A thr ee-dimensional gradient-echo technique was implemented on a Signa clin ical imaging system. Data from phantoms and patients were analyzed for volume using a segmentation algorithm designed with: 1) correction fo r partial volume averaging; 2) insensitivity to user bias; and 3) spee d. Accuracy, precision, and intra- and interobserver variability were determined. RESULTS: Average error for phantom studies was 4% to 6%, o r 1 to 2 cc across the volumes, which ranged from normal to mild hydro cephalus (<60 cc). Patient studies showed intra- and interobserver err or of 2.3% and 7.8%, respectively. The correction for partial volume a veraging resulted in a threefold decrease in error. Data were acquired and reconstructed within 7 minutes. Experienced radiologists required less than 15 minutes to perform each analysis. CONCLUSIONS: This algo rithm allows accurate measurement of ventricular volumes in an efficie nt, minimally supervised manner.