Expertise Related Changes in Resting-State Functional Connectivity Patterns Following a Clinical Reasoning and Decision-Making Task

Abstract

Purpose: This study investigated the behavioral and resting-state neural correlates of clinical decision-making among expert gastroenterologists and novice medical students, aiming to understand how diagnostic expertise is reflected in either pre-task and/or post-task brain activity.

Method: Participants completed a clinical decision-making task while behavioral measures (accuracy and response time) were recorded. Resting-state fMRI data were acquired immediately before and following the task. Group differences in brain connectivity were analyzed using seed-based connectivity and multivariate partial least squares (PLS) analyses, focusing on the frontopolar prefrontal cortex (FPPFC) and its associated networks.

Finding: Experts outperformed novices in diagnostic accuracy and speed, especially on "easy" cases, suggesting enhanced cognitive efficiency. Experts also showed more pronounced response time variation with task difficulty, potentially reflecting strategic modulation. Resting-state fMRI revealed that experts had increased post-task connectivity between the FPPFC and the paracingulate gyrus (PaCG), a brain area associated with the executive control network. Novices, by contrast, showed stronger FPPFC connectivity with the posterior cingulate cortex (PCC), part of the default mode network (DMN), indicating a return to internally directed cognition. PLS analyses further revealed that experts engaged executive and attentional network regions post-task, while novices primarily activated DMN regions. Notably, for the expert group only, increased brain activity in attention-related regions was associated with gastroenterologists who had slower, deliberate responses on easy cases.

Conclusion: Clinical expertise is associated with sustained engagement of goal-directed neural networks after task completion, potentially reflecting ongoing cognitive evaluation or preparation. In contrast, novices appear to disengage more readily, reverting to self-referential thought. These findings highlight distinct neural mechanisms that may support the development of diagnostic expertise.

Keywords: clinical reasoning; expertise; functional magnetic resonance imaging; resting‐state; univariate and multivariate analyses.

FIGURE 1

Group mean performance (left: accuracy; right: response times) and standard errors for Novices (dark gray) and Experts (light gray) during a clinical reasoning task while acquiring BOLD signal in an MR scanner (acquired during Hruska et al. (2016a; 2016b).

FIGURE 2

Univariate GLM group‐level mixed‐effects condition by group interaction in seed‐based connectivity. Maps show brain regions (red‐outlined circles) where post‐task > pre‐task rs‐fMRI between gastroenterologist experts (light/blue clusters) and novice second‐year medical students (yellow/orange clusters) overlaid on a standard MNI152 brain image. Abbreviations: FWE = family‐wise error, PaCG = paracingulate gyrus, PCC = posterior cingulate cortex, R = right hemisphere, x = sagittal slice, z = axial slice, z = z‐score threshold value used.

FIGURE 3

Correlations bar graph (A) and corresponding functional connectivity maps (B) resulting from seed‐PLS analysis examining the relationship between groups (Novices/Experts) and conditions (pre‐/post‐task resting‐state). The bar graph (A) indicates the correlations calculated from the first significant LV where group differences in post‐task resting‐state functional connectivity patterns with the seed region (right FPPFC) are identified in the spatial maps (B). Error bars indicate 95% confidence intervals from bootstrap estimation. Error bars that cross the x‐axis indicate unstable brain scores (i.e., not significant). The spatial maps (B) display yellow/red clusters representing brain regions with increased connectivity with the seed region in the novice group during post‐task resting‐state. The light‐blue/blue clusters represent brain regions with increased connectivity with the seed region in the expert group during post‐task resting‐state. Abbreviations: LV1 = first latent variable, Post = post‐task resting‐state condition, Pre = pre‐task resting‐state condition, R = right hemisphere, z = axial slice (MNI152 standard space).

FIGURE 4

Correlations bar graph (A) and corresponding functional connectivity maps (B) resulting from seed‐PLS analysis examining the relationship between groups (Novices/Experts) and conditions (pre‐/post‐task resting‐state). The bar graph (A) indicates the correlations calculated from the second significant LV, where group differences in pre‐ and post‐task resting‐state functional connectivity patterns with the seed region (right FPPFC) are identified in the spatial maps (B). Error bars indicate 95% confidence intervals from bootstrap estimation. Error bars that cross the x‐axis indicate unstable brain scores (i.e., not significant). The spatial maps (B) display yellow/red clusters representing brain regions with increased functional connectivity with the seed region in the novice group during the post‐task resting‐state condition. The light‐blue/blue clusters represent brain regions with increased functional connectivity with the seed region in the novice group during the pre‐task resting‐state condition and the expert group in both pre‐ and post‐task resting‐state conditions. Abbreviations: LV2 = second latent variable, Post = post‐task resting‐state condition, Pre = pre‐task resting‐state condition, R = right hemisphere, z = axial slice (MNI152 standard space).

FIGURE 5

Correlations bar graph (A), brain activity maps (B), and scatterplot (C) describing the behavior‐PLS analysis examining the relationship between performance (i.e., response times in “easy” case scenarios between groups (Novices/Experts) and the post‐task resting‐state condition. The bar graph (A) indicates the group‐ and condition‐dependent correlations calculated from the only significant LV (p < 0.04) between response time performance and brain regions during the post‐task resting‐state condition identified in (B; light‐blue/blue clusters). Vertical error bars indicate 95% confidence intervals from bootstrap estimation. Error bars that cross the x‐axis indicate unstable brain scores (i.e., not significant). Scatterplot (C) shows the correlation between the post‐task resting‐state brain activity and response time performance while making clinical decisions for “easy” case scenarios. Each black square represents the average response time for each expert gastroenterologist when answering an “easy” case scenario. Horizontal bars represent the standard error of response times. Abbreviations: r = Pearson correlation coefficient of trendline, R = right hemisphere, x = sagittal slice, z = axial slice (both in MNI152 standard space).