The Prediction of Cerebral Palsy in Infants Using Pose Estimation Techniques Within a Hybrid Deep Learning Framework

Cerebral palsy (CP) is a neurological disorder that affects movement and posture, and early diagnosis is crucial for timely interventions to improve long-term outcomes. Despite advancements in diagnostic techniques, accurate early detection remains challenging. This study aimed to investigate the effectiveness of using pose estimation techniques and hybrid deep learning (DL) approaches for predicting CP in infants. Three hybrid DL models were explored: a convolutional neural network (CNN) combined with long short-term memory (LSTM) networks (CNN-LSTM), a CNN with Gated Recurrent Units (CNN-GRU), and a Graph Convolutional Network (GCN) with GRU (GCN-GRU). To further enhance prediction accuracy, an ensemble method using majority voting was employed to aggregate the predictions from these three models to maximize their strengths while minimize individual weaknesses. Performance evaluation shows that the CNN-LSTM model achieves the highest accuracy, with 100% on training and 90% on testing datasets. The CNN-GRU model achieved a training accuracy of 96% and testing 90%, and the GCN-GRU model achieved a training accuracy of 90% and testing accuracy of 90%. These findings highlight the potential of integrating pose estimation with DL techniques to enhance early CP diagnosis, offering a promising tool for clinicians to improve predictive accuracy and enable earlier, more targeted interventions.