Premature infants respond to different stimuli. This research intends to assess the temporal and spatial convolution layers of a convolutional neural network (CNN) to enhance our understanding of their operations. Additionally, it seeks to analyze and compare the variations and similarities in electroencephalography (EEG) responses among premature infants, particularly those at 31 weeks of gestational age (wGA), when exposed to repetitive and alternating auditory stimulus patterns. In our study, we employed the CNN to classify two auditory sequences: \\\"ga ga ga ga ga\\\" and \\\"ga ga ga ga ba\\\" or \\\"ba ga ba ga ba\\\" and \\\"ba ga ba ga ga\\\". The use of the CNN is advantageous because it can effectively handle noise and does not necessitate manual feature extraction. By performing average area under the curve (AUC) calculations for each brain region via the weights and trained network, we were able to assess the discrimination capabilities of different regions. The AUCs for the CNN in distinguishing syllables from repetition and alternation stimuli were 0.91 and 0.92, respectively. In premature infants, alternating sequences generate a delayed response compared to repetitive sequences. Results clearly indicate that the right temporal frontal region has the highest AUC values for the repetition protocol, whereas the left temporal frontal region has the highest AUC values for the alternation protocol. By integrating spatial and temporal convolutions in our CNN model, we effectively captured the complex interactions between auditory processing and cognitive functions.

Keywords