Video-Based Posture Analysis for Evaluating Endotracheal Intubation Performance

Endotracheal intubation (ETI) is a critical procedure that requires effective training and assessment to ensure successful oxygen delivery. Traditional training methods, such as observation and checklist-based assessments, are resource-intensive and heavily reliant on expert supervision. This study evaluates provider posture during ETI and explores the use of video-based posture analysis and machine learning to classify successful intubation attempts.

Eighteen novice participants performed ETI on a manikin for 30 repetitions over 3 visits (10 per visit) within 1 week. Videos of the third visit were cropped to the final 5 seconds of the task and processed at 4 frames per second. Provider posture points were extracted using a pose detection network, and relative distances between points were calculated as features across a dataset of 2,429 frames. After filtering out incomplete data, the 922-frame dataset was split for 10-fold cross-validation in an XGBoost model.

The model achieved an average test performance of 77.8% accuracy, 76.9% sensitivity, and 78.7% specificity on average across folds in classifying successful and unsuccessful intubations. Feature analysis identified key spatial relationships, such as the relative positions of elbows and wrists, as significant predictors of success.

This study demonstrates the potential of video-based posture analysis in objectively evaluating ETI performance. The approach minimizes the need for expert oversight and offers a scalable solution for training in resource-constrained settings. Although additional validation is needed, this method could enhance training effectiveness in environments such as military healthcare.

Integrating posture analysis and machine learning provides a practical, scalable framework for assessing ETI performance. Future research should focus on refining this methodology and exploring its integration into current training programs.