Analyzing the walking characteristics of cerebral palsy patients from kinematic data using machine learning

Abstract

Objective: Cerebral palsy (CP) is a condition based on neuromotor impairments that can particularly affect walking kinematics. Individualized assessments are necessary to improve the movement abilities of CP patients. Measurements of ground reaction forces (GRF) used in laboratories for these assessments play a significant role in the treatment of CP patients. However, measuring GRF during natural walking is challenging and costly due to equipment requirements. This study introduces a new approach for analysing the walking of CP patients by using machine learning to estimate their GRFs without the need for force platforms. Method: The research utilized walking data from 40 healthy individuals and 40 CP patients. Initially, a model was developed using kinematic data as input into a one-dimensional convolutional neural network (CNN) to classify healthy individuals and those with CP. Subsequently, the kinematic data of CP patients were fed into a second CNN model to predict GRFs. Results: The classification of CP patients and healthy individuals was achieved with high accuracy (98%). Additionally, GRFs in CP patients could be estimated with a normalized mean squared error of 13.3% (± 5.1) and a Pearson correlation coefficient of 0.88 (± 0.07). These results are crucial for patients where GRF measurement in a laboratory setting is not feasible. Conclusion: This study presents a useful and rapid movement analysis method for the treatment of CP patients, aiming to improve their treatment processes. However, it's important to handle the results obtained from the study with careful consideration to avoid errors in clinical decision-making processes.