Article
Neuronal network analysis of A-trains during vestibular schwannoma surgery in patients with and without a separate intermedius nerve
Search Medline for
Authors
Published: | June 9, 2017 |
---|
Outline
Text
Objective: Specific pathophysiologic activity in the EMG of the facial muscles, so called “A-trains”, correlate with the degree of postoperative facial palsy after vestibular schwannoma surgery in the cerebello-pontine angle (CPA). However, it has been demonstrated that a separate intermedius nerve distinguishable from the facial nerve trunk in the CPA presents a confounder and causes a vast increase of A-train activity without correlation to the functional status. Due to the complex interplay of tumor size, potentially harmful surgical maneuvers, presence of a separate intermedius nerve, A-train activity and postoperative facial palsy, we investigated the use of a neuronal network, which is suited to capture such interrelations implicitly.
Methods: The amount of automatically detected A-train activity and tumor sizes of 144 patients who underwent first surgery for a vestibular schwannoma were used to train and evaluate a neuronal network (NN). The chosen architecture was a feedforward network with a single hidden layer. “Traintime”, the overall quantity of A-trains, of 9 channels (electrodes in the m. orb. oculi and oris, as well as in the perinasal muscles) and the tumor size (Koos classification) were used as inputs. The immediately postoperative facial nerve function (graded according to House and Brackmann, HB), as well as after approx. six month were uses as outputs. The 144 cases were automatically divided into training, validation and testing groups (70, 15, and 15%). The resulting estimated HB grades of the network were compared to the actual HB grades.
Results: A total of 70 patients presented without separate intermedius nerve. Spearman-correlation of traintime to postoperative HB without application of the NN methodology was 0.45 (p<0.0001). The 74 patients with separate intermedius nerve showed a correlation of 0.15 (p>0.1) and presented with significantly larger tumors (p<0.05). Correlation of NN-estimated with actual HB grades was 0.68 (p<0.0001).
Conclusion: Neuronal networks are able to compensate the confounding influence of a separate intermedius nerve, based on the total traintime per channel and the tumor size. The results suggest an interrelation of excess intermedius activity and tumor size.