Hemodynamic matrix factorization for functional magnetic resonance imaging

TY - GEN

T1 - Hemodynamic matrix factorization for functional magnetic resonance imaging

AU - Hütel, Michael

AU - Antonelli, Michela

AU - Ekanayake, Jinendra

AU - Ourselin, Sebastien

AU - Melbourne, Andrew

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Neural activation causes a complex change in neurophysiological parameters of the cerebral blood flow (CBF). Functional magnetic resonance imaging (fMRI) measures one of these neurophysiological parameters, which is the blood oxygen level dependent (BOLD) response. The general linear model (GLM) used in fMRI task experiments relates activated brain areas to extrinsic task stimuli. The translation of task-induced neural activation into a hemodynamic response is approximated with a convolution model in the GLM design. There are major limitations to the GLM approach. First, the GLM approach does not model intrinsic brain activity. Second, the GLM assumes compliant task participation matching the stimulus timing and duration in the corresponding task. We propose hemodynamic matrix factorization (HMF), a data-driven approach to model intrinsic and extrinsic neural activation in fMRI. By contrast to the GLM, the HMF does not incorporate the original task design. The neural activation is a latent variable and estimated from fMRI data. Each component of the HMF consists of a neural activation time course and a spatial mapping. A linear filter translates neural activation time courses into BOLD responses. We apply our HMF to a motor localization task of an open source data cohort. We obtain neural activation time courses that correlate with the original block design of the task and whose corresponding spatial maps match individual areas of the sensory-motor cortex known to be activated by either foot, hand or tongue movement. We find HMF components whose neural activation time courses correlate with the visual cue timings presented at the beginning of each task block. HMF thus constitutes a novel tool to validate if the actual task execution of a subject matches the intended execution specified in the task design of fMRI experiments.

AB - Neural activation causes a complex change in neurophysiological parameters of the cerebral blood flow (CBF). Functional magnetic resonance imaging (fMRI) measures one of these neurophysiological parameters, which is the blood oxygen level dependent (BOLD) response. The general linear model (GLM) used in fMRI task experiments relates activated brain areas to extrinsic task stimuli. The translation of task-induced neural activation into a hemodynamic response is approximated with a convolution model in the GLM design. There are major limitations to the GLM approach. First, the GLM approach does not model intrinsic brain activity. Second, the GLM assumes compliant task participation matching the stimulus timing and duration in the corresponding task. We propose hemodynamic matrix factorization (HMF), a data-driven approach to model intrinsic and extrinsic neural activation in fMRI. By contrast to the GLM, the HMF does not incorporate the original task design. The neural activation is a latent variable and estimated from fMRI data. Each component of the HMF consists of a neural activation time course and a spatial mapping. A linear filter translates neural activation time courses into BOLD responses. We apply our HMF to a motor localization task of an open source data cohort. We obtain neural activation time courses that correlate with the original block design of the task and whose corresponding spatial maps match individual areas of the sensory-motor cortex known to be activated by either foot, hand or tongue movement. We find HMF components whose neural activation time courses correlate with the visual cue timings presented at the beginning of each task block. HMF thus constitutes a novel tool to validate if the actual task execution of a subject matches the intended execution specified in the task design of fMRI experiments.

UR - http://www.scopus.com/inward/record.url?scp=85075650820&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-32391-2_12

DO - 10.1007/978-3-030-32391-2_12

M3 - Conference contribution

AN - SCOPUS:85075650820

SN - 9783030323905

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 117

EP - 125

BT - Connectomics in NeuroImaging - 3rd International Workshop, CNI 2019, Held in Conjunction with MICCAI 2019, Proceedings

A2 - Schirmer, Markus D.

A2 - Chung, Ai Wern

A2 - Venkataraman, Archana

A2 - Rekik, Islem

A2 - Kim, Minjeong

PB - SPRINGER

T2 - 3rd International Workshop on Connectomics in NeuroImaging, CNI 2019 held in conjunction with the 22nd International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2019

Y2 - 13 October 2019 through 13 October 2019

ER -