Automated Pediatric Delirium Recognition via Deep Learning-Powered Video Analysis

Area of Science:

• Artificial Intelligence
• Medical Informatics
• Computational Neuroscience

Background:

• Pediatric delirium is a common and challenging condition with a higher incidence than in adults.
• Accurate and timely recognition of pediatric delirium is crucial for effective clinical management.
• Current diagnostic methods can be time-consuming and labor-intensive for healthcare professionals.

Purpose of the Study:

• To develop and validate a deep learning-based algorithm for automated recognition of pediatric delirium from video recordings.
• To assess the performance of the proposed model in terms of accuracy, precision, recall, and F1-score.
• To enable intelligent video diagnosis for pediatric delirium within a hospital system.

Main Methods:

• Collected 129 video samples (74 non-delirium, 55 delirium) labeled by clinicians.
• Employed an 18-layer deep spatiotemporal convolutional neural network (CNN) with 2D and 1D convolutional filters.
• Pretrained the architecture on a large-scale video dataset and integrated a three-layer fully connected classification head.

Main Results:

• The proposed algorithm achieved a robust classification performance with an accuracy of 0.8718, precision of 0.8711, recall of 0.8730, and F1-score of 0.8715.
• The model demonstrated clinical applicability and technical reliability across various training and testing strategies.
• An independent test set of 100 new samples yielded an accuracy of 0.8800, confirming the model's effectiveness.

Conclusions:

• The developed deep learning algorithm provides a reliable and effective method for pediatric delirium recognition.
• The model's deployment in a hospital system facilitates intelligent video diagnosis, potentially improving patient outcomes.
• This approach offers a novel solution for addressing the challenges in diagnosing pediatric delirium.