Enhancing accuracy of magnetic resonance image fusion by defining a volume of interest

Abstract

We compared the registration accuracy for corresponding anatomical landmarks in two MR images after fusing the complete volume (CV) and a defined volume of interest (VOI) of both MRI data sets. We carried out contrast-enhanced T1-weighted gradient-echo and T2-weighted fast spin-echo MRI (matrix 256 x 256) in 39 cases. The CV and a defined VOI data set were each fused using prototype software. We measured and analysed the distance between 25 anatomical landmarks in predefined areas identified at levels L(1)-L(5) corresponding to defined axial sections. Fusion technique, landmark areas and level of fusion were further processed using a feed-forward neural network to calculate the difference which can be expected based on the measurements. We identified 975 landmarks for both T1- and T2-weighted images and found a significant difference in registration accuracy ( P<0.01) for all landmarks between CV (1.6+/-1.2 mm) and VOI (0.7+/-1.0 mm). From cranial (L(1)) to caudal (L(5)), mean deviations were: L(1) CV 1.5 mm, VOI 0.5 mm; L(2) CV 1.8 mm, VOI 0.4 mm; L(3) CV 1.7 mm, VOI 0.4 mm; L(4) CV 1.6 mm, VOI 0.6 mm; and L(5) CV 1.6 mm, VOI 1.6 mm. Neural network analysis predicted a higher accuracy for VOI (0.05-0.15 mm) than for CV fusion (0.9-1.6 mm). Deviations due to magnetic susceptibility changes between air and tissue seen on gradient-echo images can decrease fusion accuracy. Our VOI fusion technique improves image fusion accuracy to <0.5 mm by excluding areas with marked susceptibility changes.

Performance Assessment for Brain MR Imaging Registration Methods

“…6 Also, using volume size as a metric is reasonable in the context of nonlinear registration algorithms that can locally deform the image volume and change the size of labeled anatomic regions, but our affine registrations create minimal regional size changes, making volume size an inappropriate metric and further motivating the use of pinpoint, single-voxel landmarks in a manner that extends prior work. 7 Additionally, surface distance, which describes registration success by using the average distance between points on one surface to the closest points on another surface, 6 disregards whether homologous points are being compared and highlights a possible lack of spatial precision. This article targets the radiologic clinic, where patient care involves the affine registration of a single patient's images across modalities and/or techniques to assess pathology.…”

“…In a study that used a widely accepted, neurosurgical commercial package, Hoelper et al 7 placed 25 anatomic landmarks in T1 and T2 brain volumes to test the registration error for 39 patients, a rare example of a publicly available assessment for a commercial registration solution. They demonstrated that whole-brain volume registrations could have errors ranging from 0.7 to 2 mm, depending on the region of the brain, and they therefore recommended using a volume of interest to improve local registrations when a particular area was important.…”

“…They demonstrated that whole-brain volume registrations could have errors ranging from 0.7 to 2 mm, depending on the region of the brain, and they therefore recommended using a volume of interest to improve local registrations when a particular area was important. 7 Such knowledge of the behavior of a commercial product can improve its use in the clinic, but if commercial vendors do not use objective metrics to characterize thoroughly the performance of their products and then make these results available, rational choices and improvements cannot be made. This situation is to the ultimate detriment of the patient undergoing treatments that rely on these algorithms being highly accurate.…”