Cost Efficiency of fMRI Studies Using Resting-State Vs. Task-Based Functional Connectivity

Abstract

We investigate whether and how we can improve the cost efficiency of neuroimaging studies with well-tailored fMRI tasks. The comparative study is conducted using a novel network science-driven Bayesian connectome-based predictive method, which incorporates network theories in model building and substantially improves precision and robustness in imaging biomarker detection. The robustness of the method lays the foundation for identifying predictive power differentials across fMRI task conditions if such differences exist. When applied to a clinically heterogeneous transdiagnostic cohort, we find shared and distinct functional fingerprints of neuropsychological outcomes across seven fMRI conditions. For example, the emotional N-back memory task is found to be less optimal for negative emotion outcomes, and the gradual-onset continuous performance task is found to have stronger links with sensitivity and sociability outcomes than with cognitive control outcomes. Together, our results show that there are unique optimal pairings of task-based fMRI conditions and neuropsychological outcomes that should not be ignored when designing well-powered neuroimaging studies.

FIGURE 1

Comorbidity patterns among psychiatric disorders. This UpSet plot illustrates our dataset's frequency and co‐occurrence of mental health diagnoses. The left axis shows the total number of patients for each disorder, while the main plot displays the size and composition of various comorbidity combinations. Each column represents a unique combination of disorders, with connected dots below indicating the specific disorders involved. The height of each bar reflects the number of patients with that exact combination of diagnoses.

FIGURE 2

Distribution of prediction accuracy by task and behavioral category. The figure presents two bar plots summarizing prediction accuracy across different fMRI conditions and behavioral categories. Left plot: This plot presents the averaged prediction accuracy across different behavioral outcomes, demonstrating how well each fMRI condition predicts neuropsychological measures. Each bar corresponds to an fMRI condition (e.g., Rest1, Eyes, etc.), and the error bars indicate the interquartile range (25th–75th percentiles), quantifying variability in prediction performance. Right plot: This visualization depicts the averaged prediction accuracy across different fMRI conditions, summarizing the predictive performance of each neuropsychological measure (e.g., BriefSymp, Control, etc.) across all included fMRI conditions. The main bars represent the mean prediction accuracy, while the error bars indicate variability, providing insight into the consistency of predictive performance across different methods and conditions.

FIGURE 3

Covariance estimates between brain regions and sociability across various fMRI conditions. This figure visualizes the covariance estimates of 268 brain nodes across different fMRI conditions: Rest, Average, gradCPT, EN‐back, Eyes, and SST. The color scale indicates the magnitude of covariance, ranging from −0.4 (blue) to 0.4 (yellow). A lighter color indicates a higher absolute value of covariance. The visualization is generated using the ROI drawing function in BrainNetViewer.

FIGURE 4

Circle plot and brain plot corresponding to top brain nodes for each task condition. The Circle plot demonstrates the functional connectivity patterns of brain nodes, specifically highlighting the top 5 positive covariance nodes (visualized in red) and top 5 negative covariance nodes (visualized in blue) for Sociability across distinct fMRI conditions. The complementary brain plot illustrates the anatomical positions of the top 20 brain nodes. Node size quantifies the absolute magnitude of the covariance estimate, while node color encodes the directionality and magnitude of covariance, with blue indicating negative values and red indicating positive values.

FIGURE 5

Prediction accuracy across tasks and categories. (A) Boxplots show prediction accuracy distributions for different tasks within each category. Colors represent various tasks, illustrating differences in variability and central tendencies. (B) Line plot depicts average prediction accuracy by category across tasks, with each line representing a category. This highlights performance trends and variations among categories over tasks. (C) Each cell represents the mean accuracy when predicting a particular behavioral category (y label) using a particular fMRI condition (x label). Calculation of mean prediction accuracy utilizes the 5 replications of all behavioral variables in each behavioral category.

FIGURE 6

Spidar plot of functional biomarkers for behavioral categories. This figure displays counts of functional biomarkers across different behavioral categories: Control, Sociability, Distress, Sensitivity, BriefSymp, Empathy, NegEmo, and PosEmo. Each plot represents the distribution of various functional biomarkers in specific anatomical regions, with different colored lines indicating different fMRI conditions. The number of functional biomarkers is derived from the functional labels assigned to the top 20 brain nodes. The top 20 brain nodes refer to the top 10 brain nodes with the largest positive covariance and the top 10 brain nodes with the smallest negative covariance. The covariance estimates between 268 brain nodes and behavior categories are calculated by the latentSNA model.

FIGURE 7

Spider plot of functional biomarkers for each fMRI condition. This figure displays counts of functional biomarkers of top 20 brain nodes (The definition of top 20 brain nodes is consistent with that specified in Figure 6) across different fMRI conditions: Rest, Eyes, gradCPT, EN‐back, SST, and Average. Each plot represents the distribution of various functional biomarkers, with different colored lines indicating different behavioral categories, including Control, Sociability, Distress, Sensitivity, BriefSymp, Empathy, NegEmo, and PosEmo.