A Machine Learning Algorithm for Prediction of Facial Nerve Injury Following Microsurgical Resection of Vestibular Schwannoma

Sabrina M. Heman-Ackah; Rachel Blue; Alexandra E. Quimby; Tiffany Hwa; Jason Brant; Michael J. Ruckenstein; Douglas C. Bigelow; Selena E. Briggs; Yale Cohen; John Y. Lee

doi:10.1055/s-0043-1762122

Journal of Neurological Surgery Part B: Skull Base, Inhaltsverzeichnis

J Neurol Surg B Skull Base 2023; 84(S 01): S1-S344
DOI: 10.1055/s-0043-1762122

Presentation Abstracts

Oral Abstracts

A Machine Learning Algorithm for Prediction of Facial Nerve Injury Following Microsurgical Resection of Vestibular Schwannoma

Authors

Sabrina M. Heman-Ackah

¹Department of Neurosurgery, University of Pennsylvania, Philadelphia, Pennsylvania, United States
Rachel Blue

¹Department of Neurosurgery, University of Pennsylvania, Philadelphia, Pennsylvania, United States
Alexandra E. Quimby

²Department of Otorhinolaryngology, University of Pennsylvania, Philadelphia, Pennsylvania, United States
Tiffany Hwa

³Department of Otolaryngology - Head and Neck Surgery, Temple University, Philadelphia, Pennsylvania, United States
Jason Brant

²Department of Otorhinolaryngology, University of Pennsylvania, Philadelphia, Pennsylvania, United States
Michael J. Ruckenstein

²Department of Otorhinolaryngology, University of Pennsylvania, Philadelphia, Pennsylvania, United States
Douglas C. Bigelow

²Department of Otorhinolaryngology, University of Pennsylvania, Philadelphia, Pennsylvania, United States
Selena E. Briggs

⁴Department of Otolaryngology, MedStar Washington Hospital Center, Washington, Dist. of Columbia, United States
Yale Cohen

⁵Department of Bioengineering, University of Pennsylvania, Philadelphia, Pennsylvania, United States
John Y. Lee

¹Department of Neurosurgery, University of Pennsylvania, Philadelphia, Pennsylvania, United States

Abstract

Volltext

Introduction: Microsurgical resection of vestibular schwannomas (VS) poses a significant risk of injury to cranial nerve VII (facial nerve), which can have substantial negative impact on patients’ quality of life. Facial nerve preservation is therefore an important goal of surgery. However, facial nerve outcomes remain challenging to prognosticate, largely due to a paucity of reported relevant predictive factors.

Objective: We endeavored to design a machine learning algorithm to decipher predictive factors relevant to facial nerve outcomes following microsurgical resection of VS.

Methods: We constructed a database of patient-, tumor- and surgery-specific features via retrospective chart review of 138 consecutive patients who underwent microsurgical resection of VS between 2018 and 2021. This database was then used to train nonlinear supervised machine learning classifiers to predict facial nerve preservation, defined as House-Brackmann (HB) I versus facial nerve injury, defined as HB II–VI, as determined at 6-month outpatient follow-up.

Results: Among the factors evaluated, only age demonstrated a statistically significant association with facial nerve outcome by independent t-test (mean age: 51.46 functionally preserved vs. 57.35 injured, difference 5.89: years, 95% CI: 0.7727–11.0073; p = 0.0339). A random forest classifier demonstrated the highest accuracy (85.7%) amongst algorithms tested for prediction of facial nerve injury, with a sensitivity of 80% and specificity of 90.1%. We identified age, body mass index (BMI) and intracanalicular tumor extension as novel prognosticators of facial nerve injury. Importantly, these same factors were identified when the algorithm was retrained according to facial nerve status at most-recent follow-up.

Conclusion: Here, we describe the development of a machine learning algorithm to predict the likelihood of facial nerve injury following microsurgical resection of VS. In addition to itself serving as a clinical tool, the factors identified can be further developed as a clinical scoring system for prospective facial nerve prognostication in VS microsurgery.