Fig 1.
An example of an ultrasound image obtained using the portable handheld ultrasound probe of a LNCaP tumor inoculated in in the right flank of a nude mouse.
The image was obtained 37 days after inoculations. The tumor is seen as a hypoechoic mass. The figure shows a transversal view of the tumor measuring 11.3 mm in width and 6.6 mm in depth (a) and a coronal view of the tumor measuring 11.1 mm in length and 6.2 mm in depth (b).
Fig 2.
Examples of MR images and ROIs for a large tumor (left, 1.3 cm3) and a small tumor (right, 0.06 cm3). MR images are shown with histogram-equalized colormaps, and ROIs are shown as orange regions with black and white dashed borders. The large tumor is well-approximated by an ellipsoid shape, and the small tumor is less accurately represented, although this is difficult to convey in two-dimensional images.
Fig 3.
Example of tumor volumes measured with caliper, ultrasound, and MRI in one mouse treated with EBRT.
The plot illustrates the data available for the method comparison, wherein growing tumors are assessed by each of the three devices, albeit at different time points. Data imputation inside the overlapping time intervals facilitated thorough comparisons on this dataset.
Fig 4.
Comparisons of estimated tumor volumes between pairs of methods.
Data points where volumes were estimated by both methods are shown as circles whereas data where one point was estimated by linear interpolation is shown as triangles. Data derived from each mouse is color coded consistently across plots. Linear regression lines are shown as dotted lines, and the identity line is shown as a broken gray line. The dotted lines show linear regression lines, with Pearson correlation coefficients of r = 0.91, p<10−20 (caliper-MRI), r = 0.94, p<10−13 (ultrasound-MRI), and r = 0.97, p<10−40 (caliper-ultrasound). The pairwise method comparisons demonstrate that caliper tend to produce greater volume estimations compared to both ultrasound and MRI, whereas the ultrasound and MRI methods are more consistent.
Fig 5.
Distributions of paired volume estimates for paired ultrasound and caliper data (left) and distributions of volume errors in paired ultrasound and caliper measurements compared to MRI as reference (right). Data is visualized as box plots and the paired data-points are shown as scatter plots connected by lines. Whiskers indicate the 5th and 95th percentiles. Data involving only measured volumes are indicated by circles, and triangles indicate points produced by interpolation. The plot on the left shows that caliper measurements systematically yield larger volumes than those performed with ultrasound. The figure on the right shows that volume estimation by ultrasound is markedly more accurate than caliper. Indeed, caliper measurements overestimate the tumor volume by 0.21 to 0.47 cm3 in 50% of cases.
Fig 6.
Panel a and b: Comparison of paired tumor lengths (a) and widths (b) measured by caliper and ultrasound. Panel c: Comparison of caliper-approximated tumor depths (i.e., widths) against tumor depths measured by ultrasound. Linear regressions are shown as dotted lines, with Pearson correlation coefficients of r = 0.91, p = <10−26 (a), r = 0.89, p<10−24 (b), r = 0.90, and p<10−24 (c). Length measurements by caliper exceed corresponding ultrasound measurements in a large majority of cases.
Fig 7.
Comparison of tumor widths and depths measured by ultrasound.
Linear regression is shown as a dotted line with a Pearson correlation coefficient of r = 0.92, p<10−28. Measurements in the ultrasound images finds that all tumors are slightly wider than they are deep, indicating that spheroid approximations may yield inaccuracies.
Table 1.
Summary of the strengths and weaknesses of the investigated methods.
When compared to caliper and MRI, 2D ultrasound offers a reasonable balance between cost, throughput, and volume accuracy.