Implementing Machine Learning to Detect Changes in Facial Nerve EMG During Vestibular Schwannoma Resection

Background: One of the most devastating complications associated with vestibular schwannoma surgery is facial nerve (FN) injury and subsequent loss of function. In order to preserve FN function surgeons have previously utilized a monopolar stimulator to assess nerve function throughout surgery. This process is time-consuming and not continuous. One potential solution to this involves the utilization of the blink reflex where FN function can be stimulated through motor potentials and monitored continuously and autonomously. Despite the potential benefit of measuring the Blink reflex, it is currently unknown if subtle changes in EMG response will be detectable by neuromonitoring technicians and the degree to which the blink reflex can predict FN function. Therefore, this study seeks to implement a machine learning model to detect key features of the blink reflex response that can predict FN function vestibular schwannoma surgery.

Methodology: To gauge the feasibility of implementing a machine learning model we selected 6 patients that underwent surgical intervention for vestibular schwannoma. During the patient’s procedure, we collected free running FN EMG data and tested the blink reflex EMG using a stimulus train of 4–7 pulses. Free running and blink reflex EMG data were loaded into a Jupyter Python 3 environment and analyzed using the MNE analysis package. Initial modeling involved visualizing the power spectrum of our blink reflex data and changes to amplitude over time.

Results: Free running and blink reflex EMG data were successfully loaded into our Python 3 environment utilizing the MNE EMG analysis package. The power spectra for the right and left blink reflex were plotted using the PSD function as defined by the MNE package ([Fig. 1A], [1B]). Additionally, we plotted changes to the amplitude of the right and left blink reflexes over time ([Fig. 2]).

Conclusion: Our initial visualizations confirm our ability to import and process large-scale EMG data using the MNE package. Additionally, our visualizations emphasize the difficulty in detecting changes to neuromonitoring data and highlight the need to implement computational solutions to identify patterns in neuromonitoring. While not impossible to detect with the human eye, machine learning models have the potential to uncover significant patterns in blink reflex response that could be indicative of changes to facial nerve function. Future work will focus on developing a specific machine learning model to detect and compare minute changes to the amplitude, frequency, and waveform in blink reflex response to FN outcomes.

Fig. 1A Left blink power spectrum.

Fig. 1B Right blink power spectrum.

Fig. 2 Left and right blink amplitude over time.

No conflict of interest has been declared by the author(s).

Publication History

Article published online:
05 February 2024