MIDMSL: A Densely-Sampled, Multimodal MRI Dataset of Motor Skill Learning

2082 

Abstract Submission 

Anna Xu¹, Ross Blair², Jaime Ali Rios¹, Rastko Ciric³, Russell Poldrack¹

¹Stanford University, Stanford, CA, ²Stanford University, San Francisco, CA, ³Stanford University, Mountain View, CA

Anna Xu  
Stanford University
Stanford, CA

Ross Blair  
Stanford University
San Francisco, CA

Jaime Ali Rios  
Stanford University
Stanford, CA

Rastko Ciric  
Stanford University
Mountain View, CA

Russell Poldrack  
Stanford University
Stanford, CA

Neural plasticity induced by motor skill learning is characterized by a functional reorganization of the motor system[1]. This reorganization has been studied via functional MRI (fMRI), which is used to image task activation changes from learning, as well as resting-state fMRI, which is used to assess functional network changes not confounded by behavioral changes[1]. However, due to experimental design limitations, previous studies leave open questions of precisely where in the brain individuals exhibit plasticity and what the exact time course of this plasticity is. Specifically, these studies typically use group-averaged brain maps, obscuring individual variability in the functional architecture of the brain, which results in less reliable network estimates and lower sensitivity to detecting more precise brain features[2]. Additionally, these studies tend to sparsely sample post-training plasticity and exclusively study task activation or resting-state network changes. This limits their ability to capture the full trajectory of motor skill acquisition and the relationship between task activation changes and functional network changes, especially at different stages of learning when the mechanism of learning can differ[1]. To overcome these limitations, we present the Multimodal Imaging Dataset of Motor Skill Learning (MIDMSL) – a densely-sampled dataset aimed at developing more complete and precise brain maps of motor skill learning by leveraging advances in precision fMRI[2] through collecting extensive amounts of data in single participants.

MIDMSL collects both task and resting-state fMRI data in single participants learning a novel, video game-like, visuomotor task (Figure 1) across repeated sessions over the course of 5 weeks. For twice a week each week, we collect 12 minutes of resting-state fMRI data and 30 minutes of task fMRI data per session. Complementing this data, we also collect T1-weighted and T2-weighted anatomical scans, fMRI data of a localizer task for each finger, and diffusion MRI data at different timepoints throughout the 5 weeks. This dataset will finalize acquisition with 10 scan sessions for 8 participants, and it will be released on OpenNeuro[3]. To demonstrate its use, we pilot analyses with one participant's data for one session to develop a map of their brain activity throughout the task. Here, we preprocessed the data using fMRIPrep[4] and ran univariate GLM analysis contrasting activity during task execution compared with rest using fitlins[5]. We also used the fMRI data from the localizer task to draw somatotopic maps of their index finger by contrasting activity for this finger compared with rest.

Prior to scan sessions, we piloted our visuomotor task through behavioral studies acquired online (N = 10), which revealed improved motor performance across time within our experimental design (i.e., greater accuracy in more difficult levels over time). Next, using data from an example subject collected in the scanner, we show greater activation in regions in the dorsal visual stream, motor cortex, frontal regions, and cerebellum in response to the visuomotor task, compared with rest (Figure 2A). We also show a similar but more diffuse pattern of activity involving these regions for the localizer task specific to the index finger, compared with rest (Figure 2B).

In summary, we describe a dataset aimed at developing a more precise and complete map of neural plasticity induced by motor skill learning. In pilot analyses, we showed the recruitment of visuomotor regions involved in this task, as well as in a localizer task specific to the index finger. Such findings may be encouraging for future work to explore the integration of vision and motor systems throughout learning, as well as provide a basis for hypothesizing network changes occurring across sessions.

Neural Plasticity and Recovery of Function

Skill Learning

Motor Planning and Execution

Visuo-Motor Functions ¹

Databasing and Data Sharing ²

ADULTS

Cognition

FUNCTIONAL MRI

Learning

Motor

Open Data

Plasticity