Skip to main content
. 2021 Nov 9;11:21989. doi: 10.1038/s41598-021-01111-x

Table 4.

Training, validation, and test splits information for the segmentation and classification models.

Task Model Training (cases) Validation (cases) Test (cases) Cases ratio Average time points per patient Timepoint distribution (number of timepoints per number of patients) Average age (mean ± SD) Average KL (mean ± SD) Sex distribution (male/female) Total WOMAC pain scores (mean ± SD) χ2 test correlation (sex) (p-values) MANOVA one-way correlation (age|BMI) (p-values)
Segmentation Bone 57 (29) 15 (8) 30 (16) 0.520 1.01 1:100 58.4 ± 8.19 0.6 ± 1.06 49/53 2.4 ± 2.90 0.745 0.413
2:1
Cartilage 118 (114) 28 (28) 28 (28) 0.977 2.0 2:87 61.6 ± 9.93 2.3 ± 0.94 90/84 4.3 ± 3.80 0.156 1 × 10–4
Classification OA 12,634 (5402) 2558 (1111) 5926 (2530) 0.428 4.78 1:179 63.2 ± 9.17 1.3 ± 1.21 9005/12,113 2.1 ± 2.95 0.121 0.190
2:396
3:419
4:601
5:1367
6:527
7:927
Chronic pain 4029 (1324) 1257 (411) 2151 (713) 0.329 2.42 1:1103 63.9 ± 9.38 1.2 ± 1.22 3510/3927 1.5 ± 2.77 0.179 0.0848
2:771
3:509
4:345
5:192
6:104
7:43

Demographic factors were controlled by testing for statistical independence across the splits using a Pearson’s chi-squared test (χ2) for the categorical sex variable and a one-way Multivariate Analysis of Variance (MANOVA) for the joint effect of age and BMI.

Bold p-values are significant (p-value < 0.05).