Gtx 1080 ti graphic cards

Training and evaluation of the OA risk assessment models was performed on a computer running a 64-bit Linux operating system (Ubuntu 16.04) with an Intel i7 7700k quad-core CPU with 32 GB DDR3 RAM and two Nvidia GTX 1080-Ti graphic cards with 3584 CUDA cores and 11GB GDDR5X RAM. A detailed description of the training and evaluation methods used for each model is provided in the Supplemental Material.
A total of 5000 knees of the 6567 knees eligible to be included in the study were selected for model training and evaluation, with the number chosen to achieve the largest sample size consisting of near equal numbers of knees without and with pain progression. Knees without and with pain progression were randomly selected and stratified using a random data generator in TensorFlow (version 1.12, Google, Mountain View, CA) into three non-overlapping datasets for training, validation, and hold-out testing. The training dataset consisted of 4200 knees (2097 knees without and 2103 knees with pain progression), the validation dataset consisted of 300 knees (150 knees without and 150 knees with pain progression), and the hold-out testing dataset consisted of 500 knees (245 knees without and 255 knees with pain progression).