Automated detection and classification of basic shapes of newborn cry melody. (August 2018)

• Record Type: Journal Article
• Authors: Manfredi, Claudia; Bandini, Andrea; Melino, Donatella; Viellevoye, Renaud; Kalenga, Masendu; Orlandi, Silvia
• Is Part Of: Biomedical signal processing and control. Volume 45(2018)
• Page Start: 174
• Page End: 181
• Publication Date: 2018-08
• Subjects: Newborn infant cry melody -- Automated analysis and classification -- Early diagnosis of neurological impairment -- Intensive care management -- Autism spectrum disorders -- Pre-speech development; 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.2018.05.033
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A fully automated method to assess 5 melodic shapes of newborn cry is proposed to support the perceptual analysis. Classification tests were performed on synthetic and recorded signals, and provided up to 98% accuracy. Being contactless and cheap, this method is well suited for routinely clinical applications. This technique could be used for early detection of possible brain injuries or neuro-developmental disorders. Abstract: The study of newborn cry is a promising non-intrusive and cheap approach to support the early diagnosis of neurodevelopmental disorders. Specifically, cry melody, the trend of the fundamental frequency (f0) over time, could add relevant information to the acoustical analysis of infant crying. To date, the cry analysis is mainly performed by paediatricians/neurologists through a perceptual examination based on listening to the cry and visually inspecting the f0 shape. Therefore, this approach is not widespread as the procedure is operator-dependent and requires a considerable amount of time often prohibitive in daily clinical practice. This paper aims at providing a support to the perceptual analysis through a fully automated method for assessing the melodic shape of newborn cry. Cry units are detected within each recording, even of long duration, and their classification is performed according to five basic melodic shapes (falling, rising, symmetrical, plateau, and complex). The method is tested on synthesized signals and applied to recordings coming from at term healthy newborns. Results are compared to the perceptual analysis performed by trained raters with up to 98% matching. Being contact-less and cheap, this method is well suited for routinely clinical applications and could be effectively related to other clinical parameters for early detection of possible brain injuries or neuro-developmental disorders.