Frontal theta brain activity varies as a function of surgical experience and task error

Introduction

Highly skilled surgeons have the ability to monitor and rapidly adapt to changes in the environment, appropriately tune into relevant information variables, select from a large repertoire of possible sensorimotor commands and execute with a smoothness that belies their many years of training.1 2 While the majority of research on surgical performance has examined the overt behavioral characteristics of such expertise (e.g., time to task completion3 4) and subjective measures of mental workload (largely examined through post hoc surveys5), there have been very few investigations into the underlying cognitive mechanisms that mediate the ability to carry out the complex sequences of action selection and execution required for surgical practice (see ref 6 for a review).

In cognitive neuroscience, the processes involved in goal-directed attention, outcome monitoring, executing motor commands and suppressing irrelevant motor responses are clustered under the label of ‘cognitive control’ (also referred to as ‘executive function’7). One putative neural correlate of cognitive control is a pattern of oscillatory brain activity known as ‘frontal theta’—a signal that can be observed on the scalp through electroencephalography (EEG) recordings and quantified by calculating signal power in the 4–7 Hz range.8

Frontal theta is considered to be critical in performance monitoring9 10 and core to error detection10 11—the key to triggering selection and prioritization of information processing12 13 and subsequent action.14 15 The recruitment of these ‘top down’ control processes is heightened in scenarios where automatic processes are insufficient for successful adaptation to the current environment,7 with the prefrontal cortex responsible for engaging a broad network of systems involved in goal-directed actions.16–20 To our knowledge, there has been no examination into the relationship between this neural signal and surgical performance to date.

Extant theories of skill acquisition often describe a shift from deliberate to automatic action selection and execution, with requisite reductions in the working memory requirements during the performance of a highly practiced action.21 A recent unifying framework for theories of cognition and action—known as the ‘Free Energy Principle’—proposes that the neocortex (involved in higher order functions, such as sensory perception and spatial reasoning) constantly makes inferences about the world and learns from experiences through the violation of its predictions.22 Viewed in this framework, frontal theta activity could serve as both a teaching signal for the system to learn that it needs to refine its prediction for the future and simultaneously, trigger the cognitive resources required to produce adaptive control.23 A more accurate world model would require fewer behavioral adjustments and thus, a reduction in the need to recruit cognitive control.

Predicated on this theory and evidence from neuroscience, we examined whether frontal theta activity could be used to distinguish between experienced and novice dental surgery trainees on a simulated drilling task carried out on a high-fidelity virtual reality simulator. We predicted that overall, novice participants would exhibit greater task-related theta activity, reflecting greater top-down engagement of cognitive control processes relative to their more experienced counterparts. Second, given that behavioral adaptation following prediction error is a hallmark of learning, we expected a relationship between performance errors and frontal theta activity.

Materials and methods

Experimental tasks

Participants completed the experiment on a haptic virtual reality dental simulator (Simodont; Moog FCS, Nieuw Vennep, Amsterdam, The Netherlands)). The simulator has previously been shown to have construct24 25 and predictive validity26 with real-world dental tasks. In this experiment, participants were asked to use the simulator to drill holes in two virtual shapes, with task difficulty operationalized as a function of the geometric complexity of the shape. The ‘Low Difficulty’ task involved drilling a simple straight shape, while the ‘High Difficulty’ task required participants to drill out a cross-shaped object.

Each shape comprised three regions: (1) a ‘Target’—participants were instructed that they must be removed; (2) a ‘Leeway’ area—which surrounded the target region on the sides and bottom (participants were instructed to avoid removing as much of this as possible); and (3) a ‘Container’ area on the sides and bottom surrounding the leeway zone and represented as a brown-colored region that participants were instructed that they must avoid during drilling (see figure 1). The goal for participants was drill/cut 99% of the target region while minimizing drilling in the leeway and avoiding the container regions as fast as they possibly could. To avoid potentially confounding order effects, we counterbalanced the presentation of the shapes across participants. Specifically, half the participants in each group performed the Low Difficulty task first followed by the High Difficulty task and the other half performed the High Difficulty task first followed by the Low Difficulty task.

Request reuse permission

Figure 1

(A) Schematic drawing of the experimental tasks: the straight shape task is defined as a low level of difficulty while the cross shape presents a high level of difficulty. (B) Location of the electroencephalography (EEG) electrodes relative to head position. Analysis focused on channels F7, F8, AF3, AF4, F3, and F4.

Results

For behavioral performance (figure 2A), we found a significant effect of group (F(1,42)=41.18, p<0.001, η_G²=0.41). As expected,24 error rates were higher for novice participants (M=16.51, 95% CI 14.99 to 18) relative to the experienced participants (M=9.68, 95% CI 8.16 to 11.2). There was also a significant effect of task difficulty (F(1,42)=5.3, p=0.03, η_G²=0.04), with more errors occurring in the High Difficulty task (M=13.9, 95% CI 12.6 to 15.1) relative to the Low Difficulty variant (M=12.3, 95% CI 11 to 13.6). There was no interaction between task difficulty and group (F(1,42)=0.16, p=0.69, η_G²=0.001). Novice participants in the Low Difficulty condition had the highest intersubject CV score (37.71%), which reduced to 28.60% in the High Difficulty condition. Experienced participants’ CV scores also reduced from Low Difficulty (25.90%) to High Difficulty (20.98%). For intrasubject CV we found marginally more variability in the novice group (19%) relative to the expert group (16.89%).

Request reuse permission

For our EEG measure of cognitive control (figure 2B), there was a significant main effect of group (F(1,42)=12.05, p=0.001, η_G²=0.15), with frontal theta activity greater for novices (M=1.23, 95% CI 1.02 to 1.43) relative to experienced participants (M=0.72, 95% CI 0.52 to 0.93), indicating an increase in the recruitment of cognitive control. However, there was no difference across tasks (F(1,42)=2.11, p=0.15, η_G²=0.02) and no interaction (F(1,42)=0.03, p=0.87, η_G²<0.001). It is notable that for frontal theta power, we observed much larger degree of dispersion of scores across the group means compared with our behavioral error measure. The most variability was observed in the novice participants in the Low Difficulty condition (score 89.19%), which reduced to 52.24% in the High Difficulty condition. For experienced participants, the easy task had an intersubject CV score of 50.43%, which increased marginally in the High Difficulty condition (55.71%). The intrasubject CV was also larger for this measure relative to error, with a score of 32.68% for novices and 27.61% for experienced participants.

Finally, given the variability in our data, we explored whether the amount of frontal theta activity exhibited by participants could be correlated with the amount of behavioral error within each of our groups and across the two tasks. We found a significant negative correlation between frontal theta activity and behavioral error rates in the experienced group while completing the High Difficulty task (r=−0.594, n=20, p=0.0058). In other words, smaller behavioral errors were associated with greater theta activity while higher error rates were correlated with lower theta activity in this High Difficulty task for expert participants. We found no other statistically significant relationships (r’s<0.16; p’s>0.46). To examine whether this observed relationship between neural activity and task error for our experienced group in the High Difficulty task was significantly greater than the pattern found in the novice group (figure 3), we compared the magnitude of the two correlations and confirmed that the patterns were reliably different from one another (z=2.1779, p=0.0294).

Request reuse permission

Discussion

Expert surgical performance is marked by seemingly effortless, flexible behavior38 that typically manifests in smoother movements, shorter operating times and fewer errors.39 40 This behavior is the product of a distributed network of neural circuitry that takes a complex sequence of action selection and planned motor execution plan and refines over time to ensure smooth and seemingly automatic performance.41 42 However, there have been very few investigations into the neural processes linking brain and behavior in the surgical domain,43–51 with technological constraints limiting the ability to probe the neural underpinnings of surgical performance.

We took advantage of recent advances in wireless EEG technology to understand these processes in more detail. We reasoned that a specific neural signal, referred to as frontal theta, a putative marker of cognitive control, would distinguish between experienced and novice dental students. We hypothesized that novice participants would require the recruitment of more cognitive resources to carry out the task relative to their more experienced counterparts and in line with this expectation, we found an increase in frontal theta activity for the novice.

We also found that the relationship between this signal and error correction was specific only to experts when performing difficult tasks. These results were not predicted a priori, and the result appears to indicate a reversal in the relationship between theta activity and performance and as such it is worth considering the nature of this relationship in detail. Here, in contrast to the global pattern in which theta power was greater for the novices relative to the experts, we found larger theta power for experts who made fewer errors. To understand the processes underlying this phenomenon, consider the example of a learner driver stepping into the driving seat for the first time. We can imagine that our student is keen to learn and thus extremely attentive to the stimuli visible on the road ahead. However, it is also clear that there is a limit to the performance levels that could reasonably be expected of our student—irrespective of the amount of attentional resources that might be recruited for the task. In this context, making the driving conditions more hazardous is unlikely to modulate the relationship between cognitive control and performance to any reasonable degree, given that performance levels are low and attentional allocation is already high. Now contrast this with a more experienced driver, fewer attentional resources are required relative to the learner to exercise a higher level of performance. But if we heighten the task demands, say through poor weather conditions, we can reason that those who make fewer errors are also likely to be the individuals with increased allocation of attentional resources. Note that for this analogy to apply to the present results, our experienced drivers should make enough errors for sufficiently worse performance, but not to the point of task failure. Indeed, on average, our experts’ behavioral performance far exceeded that of the novice participants.

These results also speak to a more general issue of the relationship between expertise and performance. Expert surgeons are often, as in this study, considered a homogenous group and their performance benchmarked against trainee groups.52 53 The present results indicate the existence of a more nuanced perspective on the relationship between experts and performance. While on average, their performance may be better than trainees (on metrics relating to time and error), the performance of any one individual is likely to be modulated by a number of factors. While the present data do not speak to causality, they do indicate a correlation between the amount of attentional allocation and performance in experienced participants and this may be a factor to consider in future comparisons—from experimental design (eg, motivation and distraction) to measurement and statistical analysis (examining heterogeneity within expert groups).

Some limitations of the present work are worth noting: while our sample size was comparable to the majority of previous research in this area, future research with larger sample sizes that have sufficient power to test the reliability of the brain—-behavior relationships identified here would be welcome. Reproducibility is the hallmark of science and we encourage replication tests of the results reported in this experiment. It is also important to note that our sample comprised dental surgery trainees who differed in 3 years of experience. Exploring these relationships across different levels of experiences and specialties will be important in testing the generalizability of these findings.

Given the increasingly lower costs of EEG technology and the ease in which these newer wearable systems can be incorporated within surgical simulation it is worth considering the potential benefits of doing so. While the recording of EEG in real-world settings is very much in its infancy, evidence is growing on the potential utility of this approach across a range of settings that could be instructive for surgical settings. For example, EEG signals are commonly used to trigger brain–computer interface robotic devices to support movement rehabilitation following stroke.54 Central to this approach is the need for individuals to be able self-regulate brainwave frequencies—a skill that is developed through neurofeedback training. This neurofeedback approach has also been used to optimize performance in neurologically healthy populations in a variety of tasks (see ref 55 for a review) with some work on developing surgical skills.56 Examining the long-term impact of such training for surgical skill development is key to advancing the implementation of these technologies.

With EEG providing a continuous measure of neural activity with temporal precision in the order of milliseconds, the signal could also be used to monitor performance and inform the provision of training content. Preliminary work has shown theta power can track changes in mental workload in military settings.57 With the majority of surgical incidents related to cognitive limitations,58 such an approach could potentially be used to self-monitor cognitive functioning in the operating theater. In the classroom, measuring frontal theta power may present an adjunct to existing measures of learning (which primarily focus on overt behavioral end-point performance) and allow trainers to modulate task difficulty to optimize skill acquisition within a training session.

To take advantages of the potential opportunities presented by low-cost wearable EEG, a more complete understanding of the neural processes underlying surgical skill acquisition and performance is necessary. The surgical education research community could benefit greatly from more frequently integrating this measure into the data collection process.

Conclusion

In this study we show that frontal theta, a putative marker of cognitive control, can differentiate between surgical experiences. The data also indicate that frontal theta power scales with the degree of behavioral adjustment following error commission only for expert surgeons performing difficult tasks. These results point towards a more nuanced interpretation of the relationship between expertise and performance—one that may be modulated by cognitive control.