Page 95 - Read Online

Page 12 Glaser et al. Art Int Surg. 2025;5:1-15 https://dx.doi.org/10.20517/ais.2024.36

make it difficult to precisely identify the sacral endplate angle. Second, femoral head overlapping,
particularly in patients with high BMI or osteoarthritis, can obscure the precise center of the femoral head.
Third, the quality of lateral radiographs, especially in patients with wide pelvises, can result in poor
visualization of anatomical landmarks due to increased soft tissue density. Fourth, metallic implants such as
total hip replacements can create artifacts that interfere with landmark identification. These factors
compound measurement uncertainty and likely contribute to the higher error rates observed for PI across
studies. Future deep learning models should specifically address these challenges, perhaps through
specialized preprocessing steps or architectural modifications designed to better handle landmark obscurity
and anatomical variations.

As this technology continues to evolve, it is highly unlikely that it will not play a role in patient healthcare. It
is of great importance for future research to ensure adequate ethical standards, as new concepts and
technologies are often met with some resistance. Issues with accountability, transparency, and permissions
could come into question by involving deep learning in the decision-making process. Therefore, the
integration of deep learning technology should come as a complementary tool in the surgical decision-
making processes, where surgeons can potentially optimize patient care pathways and improve overall
clinical outcomes.

Limitations
This review has certain limitations. The literature search was restricted to studies published in English,
potentially excluding some relevant non-English studies. Searches were limited to four databases, although
additional sources were hand-searched. Study screening and data extraction were performed by only two
reviewers. The meta-analysis combined studies using different deep learning architectures and imaging
modalities, which may have introduced heterogeneity. Only mean absolute errors and correlation
coefficients were synthesized, although various other accuracy metrics were reported in the studies.

An additional limitation that should be taken into consideration is that the included studies did not account
for anatomic variations such as LSTV. The prevalence of LSTV varies widely within the literature, ranging
anywhere from 3.3% to 35.6%. A recent study by Khalifé et al. demonstrated that patients with low-grade
LSTV, defined as Castelvi I and II, have similar alignments as PI-matched no-LSTV and, therefore, should
have their measurements taken from S1. Patients with high-grade LSTV, defined as Castelvi III and IV, have
more kyphotic L5-S1 segments with more cranial lumbar apex and thoracolumbar inflection point and,
therefore, should have their measurements taken from L5. Future studies involving machine learning
models for measuring spinopelvic parameters may have to pre-identify patients with LSTV and manually
input the starting point to account for these anatomic variations .
[51]

Conclusion
In conclusion, the breadth of imaging, network architecture details, spine pathologies, and statistical
validation encompassed within these studies support automated measurement of spinal curvature as viable
for clinical integration pending minor reporting enhancements. Multicenter datasets and model access
could additionally reinforce external validity and enable incremental developments in this space.

Overall, this review supports deep learning as a potentially transformative technique for automated
spinopelvic measurement from radiographs pending rigorous multicenter validation. These AI technologies
may eventually improve efficiency, accuracy, and reliability for quantitative spine image analysis.