Can we trust the accuracy of the automatic calibration of the EOS system to measure lower limb length inequality after total hip arthroplasty? Comparison of EOS versus manual measurement on 110 calibrated radiographs.

Introduction: Traditionally, to determine a length on a limb radiograph after total hip arthroplasty (THA), calibration is performed manually with the diameter of a ball or the femoral head. More recently, the development of EOS with automatic calibration has called into question the usefulness of manual calibration to highlight lower limb length inequality (LLLI). However, the validation of EOS with automatic calibration without landmarks to measure length inequalities on large images has not been verified against manual measurements on calibrated radiographs (conventional method), which motivated the present work.

Hypothesis: EOS is more accurate and reproducible from one reader to another in highlighting LLLI after THA than the classic method.

Patients and methods: One hundred and ten patients included underwent primary THA surgery in 2 centers, with postoperative EOS performed. This EOS was extracted in 2 formats: a DICOM file with automatic calibration and an uncalibrated JPEG file (220 radiographic files in total). Two readers, without knowledge of the clinical data, each analyzed all of these images using 2 methods: by measuring the distance between the center of the femoral head and the center of the mortise on the operated side and the non-operated side on the DICOM file, therefore via the software integrated into the EOS image (method 1), or by measuring these same distances on the JPEG file by calibrating the measurement to the diameter of the prosthetic head (method 2). The reproducibility of the measurements from one reader to another (inter-observer reproducibility) and the agreement of the measurements between the 2 methods (inter-method agreement) were evaluated using the intraclass correlation coefficient (ICC) and the Bland- Altman graphic method.

Results: Inter-method agreement was satisfactory regardless of the reader but with a significantly higher agreement for reader 1 (inter-method agreement for reader 1: CL1 = 0.95 and for reader 2: CL2 = 0.90 (p = 0.008)) and this result was also confirmed by the Bland-Altman plot with no bias tendency for each reader and less than 5% of measurements that were outside the agreement band. Inter-observer reproducibility for method 1 was better than that for method 2 to highlight an LLLI after THA according to the CCI. Inter-observer reproducibility for method 1 (RM1) was 0.96 and 0.92 for method 2 (RM2) (p = 0.009). This result was confirmed by the Bland-Altman plot with a mean difference of less than 1 mm for each of the methods (-0.2 (standard deviation = 2.23) and 0.81 (3.03) for method 1 and method 2 respectively).

Discussion: Our hypothesis has therefore been partially verified. The use of the calibrated software integrated into the EOS system and manual calibration are two valid methods for searching for an LLLI. The measurements remain more reproducible from one reader to another with the calibrated software integrated into the EOS system.

Level of evidence: III; non-randomized prospective comparative study.