TY - GEN
T1 - Face the Needle
T2 - 18th IEEE International Conference on Automatic Face and Gesture Recognition, FG 2024
AU - Rudokaite, Judita
AU - Ertugrul, Itir Onal
AU - Ong, L. L.Sharon
AU - Janssen, Mart P.
AU - Huis In't Veld, Elisabeth
N1 - Publisher Copyright:
© 2024 IEEE.
PY - 2024
Y1 - 2024
N2 - There are physiological, hormonal and psychological markers that occur early in a procedure involving needles. These so-called vasovagal reactions range from feeling nauseous, dizzy, to completely passing out. In an early stage, they are difficult to measure and self-report before it is too late to prevent them. This study aims to explore different features from regular video and thermal facial video recordings of blood donors in the waiting room, prior to a blood donation procedure, in order to assess to what extent it is possible to predict whether a donor will experience a low or high level of vasovagal reaction later on during the blood donation. The results showed that the best performance was achieved using pre-trained ResNet152 models with GRU on a continuous video stream, achieving an F1 of 0.69, a PR-AUC score of 0.81, and an MCC score of 0.56. This model also achieved a precision of 0.52, recall of 0.94, F1 score of 0.67, and MCC score of 0.42 on new, previously unseen mobile video data. Although the model requires further improvement, it outperforms self-reported vasovagal reaction scores and shows the potential to predict who is at risk of experiencing vasovagal reactions using facial video data.
AB - There are physiological, hormonal and psychological markers that occur early in a procedure involving needles. These so-called vasovagal reactions range from feeling nauseous, dizzy, to completely passing out. In an early stage, they are difficult to measure and self-report before it is too late to prevent them. This study aims to explore different features from regular video and thermal facial video recordings of blood donors in the waiting room, prior to a blood donation procedure, in order to assess to what extent it is possible to predict whether a donor will experience a low or high level of vasovagal reaction later on during the blood donation. The results showed that the best performance was achieved using pre-trained ResNet152 models with GRU on a continuous video stream, achieving an F1 of 0.69, a PR-AUC score of 0.81, and an MCC score of 0.56. This model also achieved a precision of 0.52, recall of 0.94, F1 score of 0.67, and MCC score of 0.42 on new, previously unseen mobile video data. Although the model requires further improvement, it outperforms self-reported vasovagal reaction scores and shows the potential to predict who is at risk of experiencing vasovagal reactions using facial video data.
UR - http://www.scopus.com/inward/record.url?scp=85199431297&partnerID=8YFLogxK
U2 - 10.1109/FG59268.2024.10581999
DO - 10.1109/FG59268.2024.10581999
M3 - Conference contribution
AN - SCOPUS:85199431297
T3 - 2024 IEEE 18th International Conference on Automatic Face and Gesture Recognition, FG 2024
BT - 2024 IEEE 18th International Conference on Automatic Face and Gesture Recognition, FG 2024
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 27 May 2024 through 31 May 2024
ER -