Evaluation of a deep learning software for automated measurements on full-leg standing radiographs

Table 4 Results of the Bland–Altman analyses for each measurement of interest

Lengths and angles	Sample size (N)	Bias [95% CI]	Lower LOA [95% CI]	Upper LOA [95% CI]
HKA (°)	331	0.19 [0.15, 0.23]	−0.89 [−0.95, −0.83]	1.27 [1.21, 1.33]
Pelvic obliquity (mm)	150	0.20 [0.019, 0.38]	−1.99 [−2.30, −1.68]	2.39 [2.08, 2.69]
Top leg length (mm)	309	0.004 [−0.19, 0.20]	−4.59 [−4.85, −4.33]	4.60 [4.33, 4.86]
Center leg length (mm)	331	0.90 [0.68, 1.11]	−4.25 [−4.53, −3.97]	6.04 [5.76, 6.32]
Top femoral length (mm)	312	−0.41 [−0.57, −0.25]	−4.53 [−4.76, −4.29]	3.71 [3.47, 3.94]
Center femoral length (mm)	334	0.50 [0.32, 0.69]	−4.45 [−4.72, −4.17]	5.45 [5.18, 5.72]
Tibial length (mm)	344	−0.087 [−0.36, 0.18]	−5.23 [−5.51, −4.95]	5.06 [4.78, 5.33]

For all measurements but pelvic obliquity, a mixed-effects approach was applied to account for data dependencies. Patient was modeled as a random effect, while radiograph and laterality of the measurement were treated as fixed effects. Sample size (N), bias, lower limit of agreement (LOA), upper LOA and their 95% confidence intervals (CI) are displayed

ISSN: 2234-2451