TL;DR
This paper introduces and validates computer-assisted algorithms (FCM and RG) for measuring the nucleus pulposus to cross-sectional area (NP-to-CSA) ratio of lumbar intervertebral discs (IVDs) from MRI scans. The study demonstrates that these automated methods are reliable and offer a more objective alternative to manual measurements, highlighting significant differences in NP-to-CSA ratios between manual and automated approaches, which is crucial for accurate biomechanical modeling of the spine.
What Problem Does This Paper Address?
Accurate characterization of the lumbar intervertebral discs (IVDs) is essential for developing robust finite element models (FEMs) that simulate spinal biomechanics and predict failure modes. While external dimensions of IVDs are relatively easy to measure, their internal composition, particularly the ratio of the nucleus pulposus (NP) to the total cross-sectional area (CSA), has been poorly investigated. Existing data on this crucial parameter is scarce, often derived from autopsy studies with potential post-mortem artifacts, and lacks detailed methodological descriptions or reliability assessments. This gap in knowledge hinders the development of precise FEMs, especially considering the significant variability in spinal geometry across individuals. Furthermore, the influence of factors like gender and spinal level on these internal dimensions has not been thoroughly explored.
Methods
The study employed two primary computer-assisted segmentation algorithms: Fuzzy C-Means (FCM) and Region Growing (RG).
- Fuzzy C-Means (FCM): An iterative clustering method that improves upon traditional C-means by allowing data points to belong to multiple clusters with varying degrees of membership. This approach is beneficial for medical image processing, especially in the presence of noise, and aims to achieve more accurate segmentation.
- Region Growing (RG): A segmentation technique that starts with "seed" pixels and iteratively adds neighboring pixels that share similar properties, forming regions. For NP segmentation, seed pixels were strategically chosen near the NP boundary due to its blurred and lower gray value compared to the center.
To validate these algorithms, manual measurements were performed by a spine surgeon on a subset of the data. The study also investigated the influence of gender and spinal level on the NP-to-CSA ratios.
Datasets
The study utilized anonymized MRI scans from a database comprising data from two previous studies:
- Archived Medical Records (AMRs): 87 subjects (44 F, 43 M) aged 20-40 years, scanned on a 1.5 T Siemens MAGNETOM Avanto.
- Asymptomatic Subjects (ASYS): 35 subjects (13 F, 22 M) with no current back disorder symptoms, scanned on a 70 cm Open Bore 3 T Siemens MAGNETOM Verio.
After applying exclusion criteria (e.g., Pfirrmann grades IV and V indicating significant degeneration), a total of 123 subjects (58 F, 65 M) and 400 lumbar IVDs were included in the final analysis.
Key Insights and Ideas
- Manual measurement variability: Manual tracing of IVD contours, even by an experienced surgeon, showed significant differences between repeated measurements and had a high absolute error, suggesting it's not a perfectly reliable method for internal composition.
- Computer-assisted methods as a reliable alternative: FCM and RG algorithms demonstrated excellent agreement with manual measurements and good agreement with each other, providing a more objective and consistent approach to quantifying NP-to-CSA ratios.
- Significant differences in NP-to-CSA ratios: Manual measurements yielded a significantly higher NP-to-CSA ratio (46%) compared to FCM (39%) and RG (38%), indicating that current FEM models might be overestimating the NP contribution if relying solely on manual data or older literature.
- Gender and spinal level influence: Both FCM and RG analyses revealed significant influences of gender and spinal level on NP-to-CSA ratios, with males generally showing higher ratios in certain lumbar regions. This highlights the need to account for these demographic factors in biomechanical modeling.
Key Results
- NP-to-CSA Ratios:
- Manual Measurement: 46% ± 6%
- FCM: 39% ± 6%
- RG: 38% ± 7%
- Manual ratios were significantly larger than FCM (7.1% difference, p=0.003) and RG (7.7% difference, p=0.001).
- FCM and RG results were not significantly different from each other (p=0.960).
- Correlation analyses showed excellent agreement between manual and FCM (PCC=0.845) and good agreement between manual and RG (PCC=0.549).
- FCM and RG exhibited excellent agreement with each other (PCC=0.837).
- Gender significantly influenced NP-to-CSA ratios (p < 0.001 for FCM, p < 0.05 for RG).
- Spinal level also significantly influenced ratios (p < 0.05 for FCM, p < 0.05 for RG).
- Interactions between gender and spinal level were also significant.
Limitations and Future Work
- Sample Size and Diversity: The study acknowledges a relatively small sample size and a lack of diversity in age and body size. Future work should include larger, more diverse populations and consider matched study designs (e.g., matched for height, age).
- Age Influence: The screening criteria for age (to ensure general health of discs) limited the ability to investigate the influence of age on NP-to-CSA ratios.
- MRI Protocol Parameters: The study did not investigate the potential influence of varying MRI scanning protocols (e.g., section thickness) on the performance of the computer-assisted methods.
- Algorithm Discrepancies: While FCM and RG showed high agreement, slight differences in statistical interpretations regarding cephalocaudal changes suggest that future investigations should explore more algorithms and potentially apply Bonferroni corrections for multiple comparisons.
- Future Directions: The authors suggest that future studies should include larger sample sizes with greater variability in age and body size. They also emphasize the importance of considering the variability of spinal morphology and internal composition in biomechanical models.
Summary updated