Table 2 Comparison of gait impairment severity predictions between the model and three clinical experts

From: Deep learning-enabled accurate assessment of gait impairments in Parkinson’s disease using smartphone videos

Examiner

Precision

Recall

Specificity

F1 score

Expert 1

0.926

0.892

0.961

0.904

Expert 2

0.761

0.692

0.833

0.668

Expert 3

0.897

0.886

0.939

0.885

Expert-Average

0.861

0.823

0.911

0.819

AI Model

0.804

0.811

0.898

0.806

  1. Evaluation with an independent test dataset of 25 participants. Performance was evaluated using macro F1 score, precision, recall, and specificity. The consensus of three clinical specialists on the UPDRS scores of participants’ gaits served as the gold standard (Table 1). Expert-Average represents the average value of each performance metric across three experts. The F1 score shows a comprehensive performance of the predictions.
  2. Bold values indicate the performance of our model.
Back to article page