Abstract

Cumulative discomfort of preterm infants can lead to abnormal development. An automated video-based discomfort detection system is proposed by analyzing motion patterns. We employ optical flow to estimate body motion across video-frames. Log Mel-spectrogram, Mel Frequency Cepstral Coefficients, and Spectral Subband Centroid Frequency features are calculated from 1D motion signals. These features enable 1D motion signals to be represented by 2D time-frequency images. Finally, deep CNNs are used on the 2D images for comfort/discomfort classification. The model was evaluated using leave-one-infant-out cross-validation on 183 video segments recorded during 17 heel prick events. Experimental results showed an AUC value of 0.985.
Original languageEnglish
Title of host publicationSPIE Medical Imaging
PublisherSPIE
Number of pages8
Volume11314
DOIs
Publication statusPublished - 2020
Event2020 SPIE Medical Imaging: Image-Guided Procedures, Robotic Interventions, and Modeling - Houston, United States
Duration: 16 Feb 202019 Feb 2020

Conference

Conference2020 SPIE Medical Imaging
CountryUnited States
CityHouston
Period16/02/2019/02/20

    Fingerprint

Bibliographical note

Conference 11314

Cite this