Size estimation and magnification error in radiographic imaging: implications for classification of arteriovenous malformations.

PURPOSE To assess magnification error in digital subtraction angiography as it pertains to arteriovenous malformation (AVM) size. METHODS A rectangular grid phantom with equally spaced markers mounted in a stereotactic frame was imaged with digital angiographic equipment. The location and orientation of the grid was altered relative to the central plane of the phantom. Both linear and area measurements were made according to the perceived location of phantom markers using a standard catheter calibration technique and compared with stereotactically derived estimates. Finally, a single case example of an angiographically imaged rolandic AVM was used to compare linear dimensions obtained with both described techniques. RESULTS The determination of location and size with standard angiographic imaging is subject to error because of the divergent geometry of the incident x-ray beam. The resulting nonconstant geometric magnification causes errors in linear measurements of 10% to 13% at depths of 7 cm from the calibration plane. Errors in area measurements at the same position increase by 20% to 25%. Measurements of maximum diameter or cross-sectional area may have an additional error when nonspherical objects are inclined to the viewing direction (40% at 45 degrees inclination). These errors are reduced to less than 1 mm using the stereotactic technique. Some commercial angiographic systems have internal software to enable a spatial calibration based on known distances in the image or on the diameter of a catheter. The catheter technique was accurate in the calibration direction (perpendicular to the catheter axis) but had a 12% error in the direction parallel to the catheter because of a nonunity aspect ratio in the video system. Measurement of the dimensions of a rolandic AVM using the catheter calibration technique had an error that ranged from -3% to +26% (standard error, 20%) with respect to the stereotactic technique. CONCLUSIONS Numerous nonstereotactic referential systems for determining linear distances are inherently erroneous by varying degrees compared with the stereotactic technique. Area and volume determinations naturally increase this error further. To the extent that no standardized method for determining linear distances exists, significant variations in estimation of AVM size result. Classification schemes for AVMs have been hampered by this technical error.