An Artificial Neural Network Model for Predicting Postoperative Facial Nerve Outcome after Vestibular Schwannoma Surgery

Introduction: Predicting facial nerve (FN) outcomes after vestibular schwannoma (VS) surgery is paramount for patient management. It holds prognostic significance and plays a crucial role in the intraoperative decision-making process, often guiding choices between gross total, near-total, or subtotal resections. The emergence of machine-learning models has significantly improved the accuracy of surgical outcomes predictions. This study aims to develop and validate an artificial neural network (ANN) model for predicting facial nerve (FN) outcome after vestibular schwannoma (VS) surgery using the proximal-to-distal amplitude ratio (P/D) along with clinical variable.

Methods: This retrospective study included 71 patients who underwent VS resection between 2018 and 2022. At the end of surgery, the FN was stimulated at the brainstem (proximal) and internal acoustic meatus (distal), and the P/D was calculated. Postoperative FN function was assessed using the House-Brackmann (HB) grading system at discharge (short-term) and after 9 to 12 months (long-term). HB grades I and II were considered good outcome, whiles grades III to VI were considered fair/poor. An ANN model was constructed, and the model was trained using 75% of the total cohort (with 10-fold cross-validation) and tested using the remaining 25%. The performance of the model was evaluated using the area under the ROC curve (AUC) for internal validation; and accuracy, sensitivity, specificity, positive and negative predictive values for external validation.

Results: The short-term FN outcome was grade I–II (good) in 57.7%, grade III–IV (fair) in 29.6%, and grade V–VI (poor) in 12.7% of patients. Initially, a model using P/D had an AUC of 0.906 (internal validation), and an accuracy of 89.1% (95% CI: 68.3–98.8%) (external validation) for predicting good vs fair/poor short-term FN outcome. The model was then refined to include only muscles with a P/D with a proximal latency between 6 and 8 msec. This improved the AUC to 0.952 and the accuracy to 100% (95% CI: 63–100%). Integrating clinical variables (patient’s age, tumor size, and preoperative HB grade) in addition to P/D into the model did not significantly improve the predictive value. A model was then created to predict the long-term FN outcome using P/D amplitude ratios with latencies between 6 and 8 msec and had an accuracy of 90.9% (95% CI: 58.7–99.8%).

Using the “weights” method, we assessed the significance of the P/D amplitude ratio calculated from the five FN innervated muscles in the prediction of postoperative FN outcome using the ANN model. The relative weights for the P/D amplitude ratio calculated from the frontalis, orbicularis oculi, nasalis, orbicularis oris, and mentalis muscles were 8.3, 0, 25.7, 26.8, and 39.2%, respectively.

Conclusions: ANN models incorporating P/D can be a valuable tool for predicting FN outcome after VS surgery. Refining the model by using a pre-processing layer based on P/D amplitude ratio calculated from CMAPs having a latency between 6 and 8 msec further improves the model’s prediction. Finally, a user-friendly interface is provided to facilitate the implementation and wider application of this model (https://emoryskullbase.shinyapps.io/postoperative_FN_prediction/).

Publikationsverlauf

Artikel online veröffentlicht:
05. Februar 2024