Table 2 Data used to train 3D DESS MRI and 2D radiograph architectures. After exclusion criteria were applied, 35,482 qualifying entries were found in the OAI dataset across 4,790 unique patients, all of which were split into training, validation, and test sets as displayed in table. To prevent data leakage, all entries from any given patient were only allowed to be in one of the three sets. S.d. is reported for age, BMI, and KOOS pain score within the table.
From: Deep Learning Predicts Total Knee Replacement from Magnetic Resonance Images
Dataset | Data type | Age | BMI | KOOS pain | Male | Female | OA status | Total entries | Unique patients | ||
|---|---|---|---|---|---|---|---|---|---|---|---|
None | Moderate | Severe | |||||||||
Training | Control | 62.5 ± 9.15 | 28.3 ± 4.75 | 87.2 ± 16.2 | 9,708 | 12,731 | 12,721 | 8,950 | 768 | 23,126 | 3,114 |
Case | 66.3 ± 8.38 | 29.6 ± 4.79 | 67.2 ± 19.6 | 291 | 396 | 41 | 357 | 289 | |||
Validation | Control | 62.4 ± 9.21 | 28.4 ± 4.64 | 87.6 ± 15.9 | 2,876 | 4,035 | 4,118 | 2,611 | 182 | 7,115 | 957 |
Case | 66.1 ± 8.76 | 29.8 ± 4.61 | 66.2 ± 19.1 | 70 | 134 | 13 | 93 | 98 | |||
Test | Control | 62.8 ± 9.55 | 28.4 ± 4.81 | 87.4 ± 16.5 | 2,126 | 2,963 | 2,892 | 2,056 | 141 | 5,241 | 719 |
Case | 66.4 ± 7.78 | 29.9 ± 3.96 | 68.7 ± 20.6 | 59 | 93 | 12 | 83 | 57 | |||
