Skip to main content

Table 1 Neonatal sleep and wake classification results (five-fold cross-validation) using different pre-trained CNNs combined with an SVM classifier

From: Can pre-trained convolutional neural networks be directly used as a feature extractor for video-based neonatal sleep and wake classification?

Video frame    Se % Sp % Ac % P %
SVM
RGB VGG16 FCL6 52.4 64.05 58.3 57.7
FCL7 52.6 75.1 64.3 66.5
FCL8 40.9 90.0 66.2 40.2
Thermal FCL6 73.1 40.16 58.4 60.2
FCL7 72.4 40.3 58.1 60.0
FCL8 71.2 41.5 58.0 60.1
RGB AlexNet FCL7 69.8 61.0 65.3 62.7
Thermal FCL7 61.0 40.4 49.9 55.0
RGB FCL8 59.5 71.2 65.7 64.7
Thermal FCL8 56.5 60.0 58.4 52.7
RGB VGG19 FCL6 81.9 36.4 52.6 54.6
FCL7 81.0 36.1 51.9 54.5
FCL8 40.9 90.0 65.2 79.4
Thermal FCL6 63.6 54.2 57.1 63.2
FCL7 67.6 45.0 54.6 60.3
FCL8 62.1 52.4 58.1 61.8
RGB InceptionV3 FCL 97.4 30.0 55.1 52.5
Thermal FCL 67.3 47.6 58.7 61.4
RGB ResNet-18 FCL 73.2 50.8 61.2 58.2
Thermal FCL 66.7 46.7 58.3 60.8
RGB GoogLeNet FCL 77.7 41.8 55.3 55.5
Thermal FCL 66.6 46.7 57.83 60.8
  1. *true positive (TP) = VEEG depict sleep and correctly identified as sleep by our feature extraction approach, false positive (FP) = VEEG depict awake and incorrectly identified as sleep by our feature extraction approach, true negative (TN) = VEEG depict awake and correctly as awake identified our feature extraction approach, false negative (FN) = VEEG depict sleep and incorrectly identified as awake by our feature extraction approach
Back to article page

BMC Research Notes

ISSN: 1756-0500

Contact us