Does Artificial Intelligence Outperform Humans Using Fluoroscopic-Assisted Computer Navigation for Total Hip Arthroplasty?

Abstract

Background

Successful total hip arthroplasty (THA) relies on the correct implant position. THA accuracy can be improved with the use of intraoperative fluoroscopic-assisted computer navigation. Artificial intelligence (AI) software may enhance fluoroscopic navigation; however, the accuracy of the AI compared to human-controlled software in assessing acetabular component position and leg length discrepancy (LLD) has not been studied.

Methods

We analyzed 420 consecutive primary THAs performed by a single surgeon using fluoroscopic-assisted computer navigation software. The first cohort of 211 patients required inputs from a human technician (manual), while the second cohort of 209 patients used an automated version of the software controlled by AI. The intraoperative acetabular component placement (inclination and anteversion) and LLD were recorded and compared to the 2-week postoperative standing anterior-posterior pelvis radiograph.

Results

Ninety-four percent (199/211) of cups in the manual cohort and 95% (198/209) of cups in the AI cohort were within the Lewinnek “safe-zone” (P = 1.0). In the manual cohort, 69% (146/211) of THAs had a final LLD within ±2 mm of the intraoperatively navigated LLD (ie, ΔLLD ≤2 mm). In the AI cohort, 66% (137/209) of THAs had a final LLD within ±2 mm of the intraoperatively navigated LLD (P = .47). Ninety-nine percent (209/211) of hips in the manual cohort and 98% (205/209) of hips in the AI cohort had a final LLD within ±5 mm of the intraoperatively navigated LLD (P = .45).

Conclusions

Both AI and human-controlled versions of the same navigation platform were similarly accurate for navigating cup position within the Lewinnek “safe zone” and LLD accuracy.