Introduction: Advancements within the field of medicine revolve around increasing the efficiency of diagnosing and subsequently treating patients. One such advancement is measurements of the central canal using artificial intelligence (AI). The authors propose the possibility of AI measuring two linear distances followed by a subsequent approximation via an area equation. The lumbar spinal canal was approximated by an area calculation using the interpedicular distance (IPD) and anteroposterior diameter (AP diameter). The three shapes evaluated were an ellipse, triangle, and rectangle.
Methods: IPD, AP diameter, and spinal canal area from L1-L5 were measured in 555 patients using the IMPAX6 (Mortsel, Belgium: Agfa-Gevaert) picture archiving and communication system. Subsequently, an approximated area of the lumbar spinal canal, assuming an ellipse shape, was calculated using ellipse equation/approximation. Triangular and rectangular approximations were done using triangle equation/approximation and rectangle equation/approximation, respectively. The equations used are the geometric equations for the area of each shape described. For example, the triangular approximation used the IPD as the base of the triangle and the AP diameter as the height. Thus, the area approximation was calculated by half of the IPD times the AP diameter.
Results: The percent error of the ellipse approximation was the lowest with a range of error from 8.44% at L1 to 15.51% at L5. The triangle approximation again was the second most accurate with a range of error starting at -26.46% at L5 to -30.96% at L1. Lastly, the percentage errors of the rectangle approximation began at 38.07% at L1 to 47.07% at L5. The ellipse and rectangle approximation consistently overestimated the area of the spinal canal, while the opposite was true for the triangle approximation. A combination of these approximations could be used to construct a second-order approximation. The approximations were all highly correlated with the authors’ manual measurements. Approximations at the L2 vertebrae were highest with a correlation of 0.934 closely followed by all approximations at L5 with a value of 0.931. Approximations were least correlated with the L4 vertebrae with a value of 0.905.
Conclusion: The correlation between the approximation equations and the measured values is significantly related. The ellipse equation best predicted the area of the spinal canal followed by the triangle and then the rectangle approximation. The percent error difference of the ellipse approximation at L1 was similar in error compared to other causes of measurement error. Continued investigation into a second-order approximation may yield a more accurate approximation.