The present study evaluated the possibilities and requirements for accurate measurement of small tumour sizes in mice using MRI. A linear, highly correlated relationship between weighed and MRI measured tumour masses down to 10−2
g was found. 3D acquisition should be considered when tumour masses of 10−1
gram or less are expected, due to the relative increase of PVE. In very small tumours (10−2
g) image acquisition at high resolution (in our setup 1603
voxels) should also be considered. The increased acquisition time using high resolution is compensated for by the smaller FOV needed to cover the tumour. Short acquisition times allow either additional MR investigations on the same animal, such as determining tumour diffusion, perfusion or metabolic parameters within one MR session, or a higher animal throughput.
Initially, the 2D method was included only for anatomical reference since gradient performance limited the minimum slice thickness to 700
μm, resulting in substantial PVE. However, the accuracy of volume estimations based on the 2D images was similar to that based on the 3D images (Figures and b). This might be due to the possibility to study adjacent image slices, which probably improves the view of the tumour shape and delineation of the tumour border. However, the 2D method has a higher interobserver variability (Table ), which supports the use of a 3D method for very small tumours, in order to limit the subjectivity in the evaluation.
In the 3D method, the turbo factor (Tf) was adjusted to reduce acquisition time. An increase in Tf results in an increased point spread function (PSF). Computer simulations assuming similar acquisition parameters as those used, and T2 values common at 7
] showed that the PSF was broadened only by a factor of 1.6 compared to the value for Tf
1 (data not shown). The minimum TEeff
is also affected by the Tf, i.e. the image contrast will vary slightly with Tf. However, tumours were always easily visualized and, altogether, the range of Tfs used in the study might only affect the results to a minor extent.
The most time consuming process in the volume determination was probably when adjusting the threshold value in images where the global segmentation had failed. To reduce the time of analysis one could e.g. calculate an average volume based on two extreme segmentations; one including most of the border, and one excluding it. Such a procedure would, however, overestimate the volume, especially for small voxel-to-diameter ratios, where asymmetry between over- and underestimated volume errors is more pronounced (Table ). In situations when small tumours require polygon delineation of the tumour border, the decision to assign voxels intersected by the polygon line to the tumour or the background compartment, will require asymmetry consideration since it might have a significant effect on the volume estimation (Table ).
The tumour density assumption (1.0
) might be an underestimation that would account for the fact that the relations between the determined masses (m3D-160, m2D
) and the weight (mT
) were less than unity. Another contributing factor could be the inherent uncertainty in the digital balance, since errors in the predictor used in regression analysis are typically manifested as a decrease of the slope coefficient towards zero [24
]. These two considerations are justified by the fact that the slope was 1.01 when m3D-160
were compared directly to each other, thus excluding the density effect and predictor uncertainties.
Few studies were found in the literature where the accuracy of MRI based tumour size measurement was verified by e.g. comparison with weight after resection. He et al. found a correlation of R2
7) when comparing volumes of pancreatic tumours in mice from T2-weighted 2D MR images (similar to our 2D method), acquired at 4.7
T, for 0.2–2.0
g tumours [6
]. We obtained a similar correlation (R2
9) but for smaller tumours (0.01–0.2
g) since we used six times smaller voxel volume, i.e. the influence of PVE might be comparable. Other groups have reported MRI tumour size measurements in mice but without verification with tumour weight, e.g. [7
]. One group reported MRI measurements of tumours in mouse pancreas down to 0.14
g at 7.0
T using sequence parameters similar to our 2D method (T2-weighted RARE sequence, 0.015
voxels), but verified the volume determination by one phantom measurement only [9
Using high resolution microCT (7
min acquisition time, voxel size of 813
, voxel volume of 0.0005
) a close correlation was found for 0.02–0.25
s.c. tumours in mice, verified by weight after resection (R2
]. Thus, microCT is a fast and accurate method, but the absorbed dose delivered to the animal, and especially to the tumour tissue, is a confounding factor in therapy response assessments.
The generally faster T1-weighted sequences have also been used for tumour imaging (e.g. [8
]). Often however, the tumour and surrounding tissue have similar T1-values, thus requiring use of contrast agents. T1-weighted sequences without contrast agents may be useful for imaging tumours in e.g. the bladder wall, where fluid constitutes the surrounding tissue, due to widely different T1-values in fluid and solid tissue [7