Table 2.

Performances of the deep-learning model.

ML model	FAC prediction model
Sample size (patients)	154 for training, 67 for validation, total 221
Sample size(images)	462 for training, 201 for validation, total 663
Sample zero ratio	Train 54.55%, validation 53.73%
CNN model details	- Mobilenet V1 with fine tuning Binary classification with sigmoid activation SGD(Stochastic Gradient Descent) optimizer, elu activation, batch size 32 Dropout regularization - Training accuracy: 73.59% - Validation accuracy: 71.64%
FAC classifier performance	- Validation accuracy: 76.12% - Validation AUC 0.751 with CI [0.649–0.852]

ML machine learning, FAC functional ambulation category, DNN deep neural network, SGD stochastic gradient descent, AUC area under the curve, CI confidence interval.