Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Modulation of speech production network in picture naming and word read-aloud: A graph signal processing analysis of electrocorticography data
Poster Session A - Sandbox Series, Thursday, October 24, 10:00 - 11:30 am, Great Hall 3 and 4
This poster is part of the Sandbox Series.
Hiromu Sakai1, Jion Tominaga1, Ji-Hyun Park1, Yuichi Kubota2, Takafumi Inoue1; 1Waseda University, 2Tokyo Women’s Medical University
Introduction: In spoken word production, the same sensorimotor output can be produced from the cortical network for speech production depending on the tasks. Picture naming and word read-aloud are good examples that produce the same output (for instance the sound “apple”) by performing two distinct tasks (naming the apple images or reading the letter string A-P-P-L-E). Previous research (Wu et al., 2011) reported that, though these tasks both activated left frontal-temporal language networks, picture naming elicited greater neural activation in the left dorsolateral premotor-prefrontal areas and word read-aloud elicited greater activation in the left ventral occipital-temporal areas. However, little is known about how these regions interact to produce the same output when performing different tasks. In this study, we use graph signal processing methods to examine the functional connectivity among regions in the analysis of electrocorticography (ECoG) recording data when the subject performed picture naming and word read-aloud tasks. Participant: A 68-year-old right-handed native Japanese speaker diagnosed as having a glioma in the left frontal region participated in this research. The patient was scheduled for surgery to remove the affected area at the Department of Neurosurgery at Tokyo Women’s Medical University Adachi Medical Center. Subdual electrodes were placed for pre-surgical examination 5 days before the experiment. Stimuli Presentation: 32 picture images and matching words in Japanese “katakana” fonts were presented on a 19-inch LCD monitor about 60 cm in front of the patient. Each stimulus was randomly presented 8 times during two experimental sessions. Each trial consists of 500 ms fixation and 500 ms stimulus presentation, followed by another 500 ms fixation. Interstimulus intervals were varied between 700 to 1300 ms. Data Acquisition: 36 grid-electrodes and 6 deep-electrodes were surgically implanted covering the left inferior to middle frontal and superior temporal to occipital-temporal areas. ECoG signals were recorded at 1,000 Hz sampling frequency by Nihon Kohden Neurofax 1200 Digital System. Analysis: Preprocessing was performed using MNE-Python analysis tools. 1,000 ms epochs were extracted from the recording data and baseline-corrected relative to 100 ms before the onset. Event-related potentials (ERPs) were calculated by averaging high frequency band activities extracted from each epoch using a band-pass filter between 65 to 155 Hz. As an in-progress analysis, Linear Predictive Coding (LPC) coefficients are obtained and use for Graph Learning (GL) to infer the underlying functional connectivity in the speech production network (Ries et al., 2020). A GL algorithm will take the entire connectivity map consisted of ECoG electrodes as graph vertices and decide the connection weights that best explain the LPC coefficients (Tavildar et al., 2019). Preliminary and Expected Results: Amplitude modulation of ERPs was observed at the electrodes placed in the left inferior frontal and temporal areas in both conditions. Picture naming elicited stronger activation at the electrodes located in the left ventrolateral frontal areas. Word read-aloud elicited stronger activation at the electrodes in the left temporal-parietal areas. GL algorithms are expected to show stronger connectivity in the dorsolateral areas for picture naming whereas in the ventral occipital-temporal areas for word read-aloud.
Topic Areas: Language Production,