Frew, A.J., Thompson, P.M., Cloughesy, T.F.,
Toga, A.W., Alger, J.R.
Biomedical Physics, Neurology, and Radiology, University of California Los Angeles, Los Angeles, CA
Parametric T2 MR imaging hypothetically offers the ability to quantitatively relate follow-up studies of individual patients to detect subtle changes in tissue properties not ascertainable from standard T2-weighted images. However, detecting changes in parametric T2 requires accurate study to study image registration so that the same volume of tissue can be compared. Automated systems for image registration and T2 calculation are available, but the question of reproducibility when they are used for longitudinal imaging of brain tumors has not been studied. The effectiveness of such automated T2 calculation and registration systems was evaluated with a longitudinal study involving 13 imaging sessions on a brain cancer patient.
A glioblastoma multiforme patient undergoing chemotherapeutic treatment without steroids was evaluated with a series of thirteen 3-Tesla imaging studies done over a period of 19 weeks prior to a surgical resection for recurrence. Each imaging study included a two echo fast spin-echo acquisition with thirty six 3 mm slices. Co-registered parametric T2 volume images at each time point were calculated from the double echo scans using a PC-based automated program, which also incorporated an automated image registration component. Registration used a 6-degree rigid body fit. Rigid body registration was confirmed by visual inspection. Tumor growth leading to mass effect and soft tissue displacement was not evident in the series of images and no local registration errors were seen in the vicinity of the tumor. Two VOIs were selected to track changes in parametric T2: A) contralateral white matter and B) the tumor volume that was resected at the end of the study. To create the contralateral white matter VOI (Fig. 1), a single T1-weighted image was registered to the parametric T2 series and then segmented into white matter, gray matter and CSF using 3 dimensional tissue classification generated by a nearest neighbor tissue segmentation algorithm. The tumor VOI was derived from the post-resection T2 weighted image. The resection cavity volume was isolated and registered to the pre-operative images using a semi-manual alignment program with a 12-degree affine fit. The VOI maps were used to mask the parametric T2 images and mean values for each VOI were calculated for the thirteen time points.
The contralateral white matter VOI (Fig. 2) showed consistent mean parametric T2 values with no distinguishable trend over the course of the 19-week study. The overall mean was 76.6 msec with a standard deviation of 3.3 msec (4.3%). During the three weeks prior to the resection, the patient suffered a recurrence which was associated with significant increase in the tumor VOI T2, as can be appreciated in the final four data points prior to resection in Fig 2. Prior to the recurrence, the overall standard deviation in tumor T2 was 4.4 msec (4%). Conclusions: Automated Image Registration procedures can be applied to longitudinally acquired parametric T2 images to detect regionally specific quantitative T2 changes in brain tumor. Longitudinal parametric T2 measurements in automatically registered images have a variance of approximately 5%.
Figure 1: Volumes of interest. A) contralateral white matter and B) tumor volume registered to pre-resection tissue location.
Figure 2: Mean Parametric T2 values in milliseconds for contralateral white matter (squares) and tumor (triangles) during the study period. The post-recurrence tumor volume T2 (open triangles) increased at a rate of 0.23 msecs/day.
Paul Thompson, Ph.D.
|RESUME| E-MAIL ME| PERSONAL HOMEPAGE| PROJECTS|