AI Analysis of Gait Shows Promise for Predicting Mobility Outcomes
-
Gait disorders are highly prevalent, especially among the elderly, and negatively impact quality of life. Quantitative gait analysis generates data to help determine illness severity and optimal treatment.
-
The aim is to use AI to analyze gait data to predict whether a patient's gait will improve at their next visit, to inform treatment decisions.
-
Raw gait cycles were fed into various ML models including LSTM, CNNs, etc. Separately, gait cycles were converted into images and fed into pretrained CNNs.
-
Best results were a validation AUC of 0.723 for an FCN model on raw signals, and 0.726 for a custom CNN on images. The FCN model performed significantly better.
-
Prediction results are encouraging given the complexity, but external validation is needed. Future work includes accounting for different pathologies and increasing the dataset size.