A Comparative Study on Two Image Analysis Methods for the Estimation of Bone Scan Sacroiliac/Sacral Ratio

INTRODUCTION

Spondyloarthritis (SpA) primarily affects the axial skeleton, leading to inflammatory and structural changes in the sacroiliac joint (SIJ), which manifest as chronic back pain, buttock pain, and stiffness.^[1,2] Early diagnosis of sacroiliitis is crucial for the timely management of patients with SpA.^[3] Conventional radiography and computed tomography have limited sensitivity for detecting early-stage sacroiliitis, while magnetic resonance imaging (MRI) can detect earlier inflammatory changes but remains costly and less accessible.^[4-6]

Skeletal scintigraphy is a sensitive and cost-effective functional imaging modality that can reveal sacroiliitis before radiographic abnormalities appear.^[7] Both the European Association of Nuclear Medicine and the British Nuclear Medicine Society endorse bone scintigraphy as a diagnostic tool for spondyloarthropathies, including ankylosing spondylitis and sacroiliitis.^[8,9] Incorporating quantitative indices, such as the sacroiliac/sacrum (SI/S) ratio, further improves diagnostic accuracy and correlation with clinical findings.^[6] However, variability in analysis techniques across institutions and software platforms has led to inconsistent results, limiting generalizability and the establishment of a universal reference database.^[10-12]

Different quantitative approaches to SI/S ratio measurement have been described. For example, some use rectangular regions of interest (ROIs) in the SIJs and sacrum, while others extract peak counts from line profiles across the SIJs and sacrum.^[4,11,12] These methodological differences can significantly alter reported values, complicating comparisons across centers or imaging systems. In addition, as sacroiliitis evolves from localized uptake in the lower joint to diffuse involvement, the choice of ROI placement and method may influence sensitivity for early or segmental disease.^[6] Previous studies have highlighted the importance of ROI definition, yet no consensus exists, and most institutions continue to establish their own normal reference ranges based on local methodology.^[4,13,14]

To address these gaps, we designed this study to compare two widely used methods of SI/S ratio calculation: the matrix method, which employs rectangular ROIs in the SIJs and sacrum, and the line method, which applies a single-pixel line segment across the SIJs and sacrum. We further evaluated interobserver and intraobserver reproducibility between experienced and novice readers. We hypothesized that methodological differences would yield significant variability in SI/S ratios and that reader expertise could affect reproducibility. Clarifying these issues may guide standardization during transitions to new software platforms and inform whether normal reference databases require adjustment when adopting different analysis methods.

MATERIAL AND METHODS

Image processing method – The matrix method

The matrix method was implemented using the I-Con p image analysis software (Siemens I-Con p, Erlangen, Germany, abbreviated I). Three rectangular ROIs – the upper, middle, and lower sections – were drawn horizontally covering both the SIJs and sacrum [Fig 1a]. A profile of the counts in the ROIs was generated. The SI/S ratios were calculated separately for each joint, specifically by dividing the maximum counts at the SIJ by the maximum counts at the sacrum, for both the right and left sides. The highest value obtained respectively in the upper, middle, and lower sections was selected as the overall representative value for that side. The left and right SI/S values were averaged to obtain the overall average value.

Close

Fig 1:

Posterior pelvis view bone scan with Tc-99m methylene diphosphonate. (a) Three long rectangular regions were drawn horizontally to cover both sacroiliac joints and the sacrum region. (b) Profile peak counts method to determine the sacroiliac/sacral ratio. Maximum counts at the peak were computed by moving an electronic cursor along the profile for the three regions.

Export to PPT

RESULTS

We found that the median SI/S ratios calculated with the line method were higher than those with the matrix method with statistical significance [Table 1]. Specifically, the overall SI/S ratios were 1.593 ± 0.290 for the line method and 1.504 ± 0.307 for the matrix method. Significant differences were observed between the two analysis methods, both in overall values and in values on subsections of both sides. For the right SI/S, the ratios were 1.592 ± 0.314 for the line method and 1.522 ± 0.346 for the matrix method, while for the left SI/S, the ratios were 1.594 ± 0.312 for the line method and 1.487 ± 0.305 for the matrix method. Subdividing the left and right SIJs into upper, middle, and lower ROIs revealed significant differences between the upper parts on the two sides and between the lower parts on the left side.

Data were analyzed separately for healthy individuals and patients with immune rheumatic diseases. In healthy subjects, the overall SI/S ratios were similar between the left and right sides and were higher with the line method than with the matrix method [Table 2]. However, in the triplicate analysis, a significant difference was only found at the lower right side. For patients with immune rheumatic diseases, the overall data presented a similar difference in results between the two methods [Table 3].

To assess interobserver variability, 3 independent observers analyzed images of 95 subjects. Their median SI/S ratios using the line method were consistently higher than those using the matrix method for all observers, but their averages were similar. Significant differences were found between the two methods for the junior observer and one of the senior observers (observer 2), but not for the other senior observer (observer 1) [Table 4].

In the intraobserver analysis, all three observers demonstrated excellent repeatability, with ICC values above 0.80 and statistically significant 95% confidence intervals [Table 5]. Similarly, interobserver agreement was high across both analysis methods, with all ICCs exceeding 0.79 and P < 0.001 [Table 6].

DISCUSSION

In this study, we demonstrated that the line method consistently produced higher SI/S ratios than the matrix method, with statistically significant differences observed in overall values and in both left and right SIJs. Both intraobserver and interobserver analyses further confirmed high reproducibility (all ICCs > 0.79, P < 0.001). These findings highlight the impact of different image processing methods on SI/S ratio quantification and underscore the need for standardization in clinical assessments.

The influence of ROI size and placement is a major determinant of quantitative results. Most hospitals practice manual analysis for pelvic scintigraphy using different equipment and software provided by various manufacturers. These often rely on either the matrix or the line method. Several studies have shown that the ROI definition significantly affects SI/S ratio outcomes.^[16-18] While Tiwari and Basu^[11] reported that ROI placement did not substantially alter indices in healthy subjects, other groups (e.g., Kaçar et al.^[4]) demonstrated that focal lesions could be overlooked when larger ROIs were averaged. Our study also showed that even among healthy subjects, significant differences persisted depending on the analysis method, supporting the importance of ROI definition in early or localized sacroiliitis.^[19]

Studies have highlighted the progressive nature of sacroiliitis, beginning in the lower third of the joint.^[7,10] Therefore,analytical methods capable of separating upper, middle, and lower joint regions may increase diagnostic sensitivity. Kaçar et al.^[4] reported patients with focal uptake in the lower joint not reflected in whole-joint averages. Our findings, consistent with Davis et al.,^[16] indicate that smaller ROIs (as used in the line method) yield higher SI/S ratios and may enhance sensitivity for early disease detection.

Another critical issue is the transition across imaging platforms and software. Davis et al.^[16] and Goldberg et al.^[12] emphasized variability between laboratories, while Kaçar et al.^[4] stressed the need for local reference values. At our institution, older databases were based on the matrix method, yet new equipment employs different algorithms.^[14]

Introducing new software raises two main challenges: (1) compatibility with existing reference ranges and (2) retraining observers.^[20] To address this, we standardized ROI placement procedures and provided observer training, which improved consistency, in line with the findings of Koç et al.^[6]

Although the line method consistently produced higher SI/S ratios than the matrix method, this study was methodological in nature and not designed to determine whether such differences affect diagnostic thresholds, clinical interpretation, or patient management. The clinical implications of methodological variability remain to be clarified and should be addressed in future research.

This study has several limitations. First, the sample size was relatively modest (n = 95). Second, only Siemens equipment was used, which may limit generalizability. Third, MRI or ASAS classification criteria were not employed as a reference standard, which restricts direct clinical validation. Future studies should include larger, multicenter cohorts, incorporate multiple vendors’ systems, and validate results against MRI or Assessment of SpondyloArthritis International Society (ASAS) standards to strengthen generalizability and clinical relevance.

CONCLUSION

This study highlights the significant differences in SI/S ratio calculations between the line method and the matrix method, emphasizing the impact of image analysis techniques on quantitative bone scintigraphy. Given these methodological differences, careful validation and adjustment of results are essential when transitioning to new image processing and analysis software. Furthermore, the feasibility of establishing a standardized normal population reference database should be reassessed to enhance consistency in clinical interpretations and diagnostic accuracy. Future research should focus on standardizing analysis methods across different imaging platforms to improve reproducibility and facilitate broader clinical application.