Article Impact Level: HIGH Data Quality: STRONG Summary of Computer Methods and Programs in Biomedicine, 250, 108195. https://doi.org/10.1016/j.cmpb.2024.108195 Guilherme C. Oliveira et al.
Points
- The study developed a computerized method using RGB videos to differentiate post-stroke patients from healthy individuals based on facial expressions, aiming to improve early stroke recognition.
- Using RGB videos from the Toronto Neuroface Dataset, the study analyzed facial expressions of 14 post-stroke patients and 11 healthy controls through orofacial examinations, employing XGBoost and regression analysis for classification.
- The method achieved 82% accuracy in distinguishing post-stroke individuals, with a sensitivity of 91% using the Kiss and Spread expressions, highlighting the importance of mouth muscle features.
- The pilot study showed promise in detecting post-stroke based on facial expressions but emphasized the need for further validation across diverse real-world settings and populations.
- The research suggests potential smartphone application integration, enabling efficient screening by first responders and facilitating timely treatment for stroke patients. Further refinement is needed to enhance accuracy and clinical applicability.
Summary
The research focuses on developing a computerized method utilizing RGB videos to differentiate post-stroke patients from healthy individuals based on facial expressions. Early stroke recognition is crucial for prompt treatment and improved outcomes, yet initial responders often miss cases. Facial expressions can serve as early indicators of stroke, prompting the investigation into using action units for facial analysis to distinguish between post-stroke individuals and healthy controls.
The study utilized RGB videos from the Toronto Neuroface Dataset, capturing facial expressions during orofacial examinations of 14 post-stroke patients and 11 healthy controls. By employing XGBoost to compute action units and regression analysis for facial expression classification, the method achieved an accuracy of 82% for distinguishing post-stroke individuals using the Kiss and Spread expressions, with a notable sensitivity of 91%. The analysis highlighted the significance of features related to mouth muscles in differentiating between the two groups.
The pilot study demonstrated the potential of the computerized method in detecting post-stroke based on specific facial expressions. While promising, further validation in diverse real-world settings and populations is essential, including individuals of various ethnicities and potential smartphone integration. The method’s viability for smartphone applications could enable first responders to conduct screening tests efficiently, facilitating timely treatment initiation for stroke patients. Further research and refinement are warranted to enhance the method’s accuracy and applicability in clinical settings.
Link to the article: https://www.sciencedirect.com/science/article/abs/pii/S0169260724001913
References Oliveira, G. C., Ngo, Q. C., Passos, L. A., Oliveira, L. S., Papa, J. P., & Kumar, D. (2024). Facial expressions to identify post-stroke: A pilot study. Computer Methods and Programs in Biomedicine, 250, 108195. https://doi.org/10.1016/j.cmpb.2024.108195
