Automatic classification of resuscitation activities on birth-asphyxiated newborns using acceleration and ECG signals. (July 2017)

• Record Type: Journal Article
• Authors: Vu, Huyen; Engan, Kjersti; Eftestøl, Trygve; Katsaggelos, Aggelos; Jatosh, Samwel; Kusulla, Simeon; Mduma, Estomih; Kidanto, Hussein; Ersdal, Hege
• Is Part Of: Biomedical signal processing and control. Volume 36(2017)
• Page Start: 20
• Page End: 26
• Publication Date: 2017-07
• Subjects: Newborns -- Birth asphyxia -- Acceleration signal -- ECG signal -- Short time energy -- Wavelet packet decomposition -- Linear discriminant analysis -- Decision tree; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2017.03.004
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Automatic classification of activities by using ECG and acceleration signals. Support the work of clinicians when observing video. Activity detection and classification system supporting health care workers. Abstract: Objectives: Newborn deaths are reported to be caused mainly by birth asphyxia. Information learned from ventilation and other treatment could help increase survival rate of newborns in need of resuscitation. Characteristics of manual bag-mask ventilation have been studied in our previous works. However, other resuscitation activities could have important impacts as well. This paper illustrates the classification of several predefined resuscitation activities using information from acceleration and ECG signal. Methods: Time and frequency domain features were extracted from the acceleration and ECG signals. A 2-stage classifier was trained on data of manually annotated activities by observing videos of 30 resuscitation babies. Leave-one-out cross validation was used: for each fold, the classifier was trained on activities of 29 patients and tested on activities of 1 patient. Results: The average accuracy of the classification of activities is 79%. Conclusions: The performance of the classification algorithms indicates that it is possible to use ECG and acceleration signals to automatically derive useful information regarding resuscitation activities.