Validation of 3D ultrasound versus MRI quantification of popliteal lymph node volume as a biomarker of inflammatory-erosive arthritis in mice

Draining lymph node (LN) enlargement has long been recognized as a hallmark of joint inflammation in rheumatoid arthritis (RA), and can be quantified with magnetic resonance imaging (MRI) (1). The importance of this outcome measure as a biomarker of inflammatory-erosive arthritis initiation, progression, and response to therapy has recently been demonstrated in murine models (2–5). Despite its potential value, this approach has not gained broad acceptance due to its very high costs (money, time and labor), and limited access to MRI machines. Thus, investigators have been evaluating ultrasound (US) as a more practical and cost-effective method to study LN in animal models (6), and RA patients (7). Based on these promising results, we aimed to validate in vivo US volume measurement of popliteal lymph nodes (PLN) in TNF-transgenic (TNF-Tg) mice (8) with varying degrees of arthritis by comparing the results obtained using both imaging modalities.

A total of 16 PLNs from 3 to 8-month-old heterozygous TNF-Tg mice (3647 line in a C57B6 background) were anesthetized with intraperitoneal ketamine (60 mg/kg) and xylazine (4mg/kg), scanned by MRI, and their volumes were determined with Amira software (Visage Imaging, Inc., San Diego, CA) as described previously (3). On the following day, these PLNs were imaged with a high-resolution small-animal ultrasound system (VisualSonics 770 with 704 scanhead). Each mouse was anesthetized with ~2% isoflurane in oxygen. Hair was removed from ankles to hips using a depilatory cream. The mouse was placed in the supine position on the 40°C heated imaging platform with paws taped to surface electrodes for heart rate monitoring and respiratory rate synchronization (Fig. 1A). The PLN was identified in brightness-mode (B-mode) by adjusting the scanhead up or down to position the PLN at the plane of focus (red arrow in Fig. 1B), and then scanned in three dimension-mode (3D-mode) with a step size of 0.032 mm. The 3D US image data were used to quantify PLN volume by manual segmentation of the lymph node and surrounding fat pads. The mean signal intensity of the fat pad (FPsi) was computed using the TissueStatistics module. To eliminate the fat pad from the node material, selected areas in which the signal intensity was over FPsi were subtracted and any resultant empty inclusions (“holes”) within the node were filled. However, holes on the surface cannot be filled by this approach. The SurfaceGen module was used to arrange the labeled pixels as a bounded surface for subsequent 3D visualization and volumetric quantification with the SurfaceView (Fig. 1C) and TissueStatistics modules. A linear regression analysis was performed on the volumes generated from both imaging modalities (Fig. 1D). We also determined the intra and inter-observer reliability of our US PLN volume measurement, which showed insignificant variability (p = 0.8399 and 0.8096 respectively).

Figure 1. Strong correlation between PLN volumes determined by MRI vs. US.

(A) Photograph of an anaesthetized mouse positioned on the heating pad with an ECG monitor, and a Scanhead 704 placed above the knee, to image the PLN with the US machine. (B) 2D US image of the PLN obtained under B-mode scan. Note the dark PLN (red arrowhead) surrounded by the bright white triangular fat pad (green arrowhead). (C) Reconstructed 3D image of the PLN (green) with surrounding soft tissue generated with US 3D-mode scan and Amira analysis software. (D) A linear regression analysis was performed by plotting the PLN volume (mm³) measured by MRI (Y-axis) versus US (X-axis). The slope and highly significant R² value are also presented. The 3D images of representative small, medium and large PLN generated independently by MRI (E, G, I) and US (F, H, J) are presented to illustrate their similarities. The colors for each PLN correspond to the color dots in D.

US proved to be a very facile approach to assess murine PLN (Fig. 1B), since they are readily identified in B-mode after locating the triangular fat pad. A strong relationship between PLN volumes measured by MRI and US was found using a linear regression model (R² = 0.844, P<0.0001) (Fig. 1D). However, vertical placement of PLN in US image is a potential source of variability, perhaps as much as 10 percent (see below), and may result from mechanical compression with the scanhead. To reduce this variation, a lower frequency scanhead could be used resulting in a greater focal depth and distance from the head. In addition, the scanhead position should be adjusted so that the PLN is centered consistently in the plane of focus (Fig. 1B). Of note is that smaller PLN are less susceptible to scanhead compression error, as suggested by the stronger linear relationship with the volume obtained from MRI. To give a broader illustration of the correlation between MRI and US measurements, 3 PLN were chosen to represent the smaller, middle and larger PLN, and their 3D rendered images generated by MRI and US are presented (Fig. 1E–J).

One surprise of our study was that the slope was not 1.0, despite the strong correlation between measurements on the same node across animals on the two instruments (Fig. 1D). To test the hypothesis that we might be compressing the node during US imaging, we varied the vertical position of the scanhead in an attempt to compress the node. In a representative test, the PLN showed a volume of 6.86mm³ at a depth of 7mm, and a volume of 6.39mm³ at a depth of 5mm, resulting in a ~10 % difference in node volume. This is not large enough to account for the differences we observed (slope = 1.45). Thus, other factors must also contribute to the differences that we observed, and accuracy of the absolute volume measurements attainable with these non-invasive approaches remains a limitation. Another error with the US measurement is the roughened surface (Figure 1F,H,J), which occurs due to our inability to fill surface holes. This should be addressable in the future with the evolution of superior surface rendering software applications. However, since the primary outcome measure of this biomarker is to predict RA progression by PLN volume enlargement, we conclude that larger PLN imaged in MRI will also be larger during ultrasound imaging. The three largest and the three smallest nodes (Figure 1D) clearly differ from one another, and the rank order of size derived from each modality is the same for these six observations. The nodes in the middle of the range are closely bunched and may not differ significantly from one another in each group.

Although palpable draining LN have long been recognized as a symptom of RA, their value as a quantitative biomarker of disease initiation, arthritic flare and response to therapy has only recently been appreciated (7). However, if this biomarker is to be broadly utilized, it needs to be assessed by practical means such as US, which can be readily performed during the office visit. For this reason, US imaging has recently been evaluated as an alternative to MRI to assess various musculoskeletal conditions. In some cases, such as detecting psoriatic arthritis of fingers and toes in patients with psoriasis (9), and detection of bone erosions in gouty arthritis (10), US has been shown to be just as effective as MRI. However, in other cases such as predicting the development of RA from undifferentiated peripheral inflammatory arthritis, MRI assessment of bone edema, synovitis and erosion pattern proved to be more useful (11).

In summary, here we demonstrate that US is comparable to MR imaging for determining relative PLN volume in mice with inflammatory arthritis. Since this can be performed at less than 10% of the financial cost, we find that US is a quick, inexpensive, and reliable method to interrogate this biomarker of RA pathogenesis and response to therapy.