Poster B122, Sunday, March 26, 8:00 – 10:00 am, Pacific Concourse
Reliability of fMRI data during speech production tasks across scanning sessions
Saul Frankford1, Alfonso Nieto-Castañón1, Frank H. Guenther1; 1Boston University
Functional neuroimaging techniques are well equipped to characterize the average neural activity for a specific speaking task compared to baseline, across healthy subjects. Average neural activity is also useful for understanding neurological diseases that are relatively homogeneous across the population. However, speech disorders that are the result of a stroke or traumatic brain injury often have characteristics unique to an individual. To map the speech production network in individuals, a crucial first step is to quantify within-subject reliability. In this study, we looked at measures of reliability in neural activation for 14 healthy subjects who participated in two similar speech production experiments between 6 and 52 days apart. We employed a classification algorithm based on principal component features to identify whether a subject’s activation pattern from one experiment most resembles the same subject in the second experiment or one of the other subjects. Using 25 principal components, the classifier maintained a 100% success rate, even when 62 additional subjects from other speech studies were included in the identification task. We also found that, on average, individual subjects had a between-session Dice coefficient of 0.73 and an intra-class correlation of 0.88. These results suggest that healthy speakers have reliable speech activation patterns from session to session and may indeed have a unique neural “fingerprint” that can be observed during speech production.
Topic Area: PERCEPTION & ACTION: Motor control