Automatic Detection of Emotional Changes Induced by Social Support Loss Using fMRI

• Published in: IEEE transactions on affective computing Vol. 14; no. 1; pp. 706 - 717
• Main Authors: Candemir, Cemre, Gonul, Ali Saffet, Selver, M. Alper
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.01.2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: BOLD signal denoising; change point detection (CPD); emotional change (EC); Feature extraction; fMRI; Functional magnetic resonance imaging; Games; Shape; Signal processing; Signal to noise ratio; social support; Task analysis; Transient analysis; Waveforms
• ISSN: 1949-3045; 1949-3045
• DOI: 10.1109/TAFFC.2021.3059965

We propose using fMRI to study emotional changes related to social-support. In this respect, a Social Support fMRI task, which triggers emotional changes was designed and implemented. The detection of emotional changes from fMRI signals has significant importance in understanding the underlying mechanisms of social-support. Unfortunately, acquired signals exhibit a very low signal-to-noise ratio and strong inter-subject variations, which render the detection process a very challenging task. For this purpose, a three-phase detection system is designed. First, possible emotional change intervals are classified to isolate trivial samples and further process the challenging ones. Second, a new denoising strategy is proposed to preserve the structural waveform properties of the emotional changes, while removing noise. Third, fMRI signals are synthesized using trapezoidal modeling and a novel feature set is extracted to characterize the varying social-support levels. The analysis shows that emotional changes can be detected automatically up to a requisite level. Despite the results cannot be generalized for the entire population due to its small sample size, our findings are meaningful and suggest further research with larger datasets. The introduced task may enable further research and the proposed system may be used as a tool for social neuroscience studies.