Cognitive biotypes identified through ECG-derived workload and behavioral accuracy

Figures

• 

  Fig. 1. Participant variability in performance, cognitive workload, and subjective experience across cognitive task difficulty levels. Panels (a) accuracy and (b) workload probability show violin plots across Easy, Medium, and Hard tasks, with red dots indicating median values and black dots representing individual participants. Panels (c) through g. display participant responses to state probes: c. perceived challenge (“How challenging was it? How hard did you have to work?”), (d) perceived success (“How successful were you in accomplishing the task goals?”), (e) stress appraisal (“How stressful, discouraging, and frustrating was the task?”), (f) mental demand (“How mentally demanding was the task?”), and (g) temporal demand (“How hurried or rushed was the pace of the task?”).

• 

  Fig. 2. Cluster Analysis of Workload and Accuracy Using K-Medoids. Three clusters of standardized workload probability and accuracy were identified using K-Medoids. Each point represents a participant, colored by cluster, and convex hulls outline cluster boundaries. Black X’s mark the medoids, and the “Avg. Dist” labels show the average distance of all points to the medoid, reflecting cluster cohesion.

• 

  Fig. 3. Cluster analysis of cognitive workload and task accuracy across participants. (a) Scatter plot of standardized cognitive workload probability versus cognitive task accuracy for all data points (3 tasks × 3 difficulty levels), colored by cluster (Cluster 1 = blue, Cluster 2 = yellow, Cluster 3 = red) with centroids (×). (b) Stacked bar chart of cluster proportions within each racial group (A = Asian, B = Black/African-American, H = Hispanic/Latino, W = White/Caucasian). (c) Stacked bar chart of cluster proportions within each gender (F = Female, M = Male). (d) Scatter plot of each subject’s mode cluster membership (most frequent cluster across nine points) with centroids. (e) Stacked bar chart of mode-cluster proportions by race. (f) Stacked bar chart of mode-cluster proportions by gender.

• 

  Fig. 4. Cluster-level discrepancies between subjective and objective measures. Panels (a-c) display box-and-whisker plots of the difference between subjective ratings and objective indices for Cluster 1 (blue), Cluster 2 (yellow) and Cluster 3 (red): (a) temporal-demand discrepancy (subjective temporal demand − ECG-based cognitive workload), (b) mental-demand discrepancy (subjective mental demand − cognitive workload), and (c) success discrepancy (subjective success − task accuracy); black lines mark the medians. Panels (d-f) show the corresponding scatter-plots: (d) workload probability vs. temporal-demand discrepancy, (e) workload probability vs. mental-demand discrepancy, and (f) task accuracy vs. success discrepancy. Clusters 1 and 3 have smaller mental- and temporal- demand discrepancies than Cluster 2, while Cluster 1 shows the largest discrepancy on the success metric.

• 

  Fig. 5. Study Design and Analysis Pipeline. The illustrated study protocol shows participants following the baseline phase to the testing phase in the same session where objective cognitive workload (ECG data) was continuously measured. In the analysis phase, task accuracy and workload metrics were clustered into three biotypes, and perceptual discrepancy was calculated as the difference between subjective state probes and objective measures. Color coding: yellow represents baseline-specific tasks; blue represents continuous objective workload monitoring; orange represents active participant behavior such as cognitive task performance and responding to subjective probes.

• 

  Fig. 6. Example second-by-second workload probability traces for three representative participants. Each panel shows the model’s continuous probability estimates of high workload (0–1) across the trial. The blue shaded region denotes the high-workload segment. In all subjects, the model exhibits a clear probability increase during the high-workload interval, demonstrating appropriate physiological reactivity.

• 

  Fig. 7. Threshold-based separation of low- and high-workload periods. Bars show the mean percentage of samples exceeding the 0.5 workload threshold during low-workload (P0) and high-workload (P1) segments. Above-threshold outputs increased from 0.042 ± 0.011 in P0 to 0.249 ± 0.073 in P1, reflecting a substantial rise in model activation during high-workload intervals.

• 

  Fig. 8. Clustering Pipeline. The figure illustrates the full pipeline for workload–accuracy clusters.