Screening Patients with Sensorineural Hearing Loss for Vestibular Schwannoma Using a Bayesian Classifier

Background: Selecting patients with asymmetrical sensorineural hearing loss for further investigation continues to pose clinical and medicolegal challenges, given the disparity between the number of symptomatic patients and the low incidence of vestibular schwannoma (VS) as the underlying cause. We developed and validated a diagnostic model using Gaussian Process Ordinal Regression, a generalization of neural networks, for detecting vestibular schwannomas from clinical and audiological data, and compared its performance with existing audiological screening protocols.

Methods: Clinical and audiometric data from 129 MR-proven VS+ and as many VS patients were obtained. A Gaussian Process Ordinal Regression Classifier (GPORC) was trained and cross-validated to classify cases as VS+ or VS−, and its diagnostic performance was assessed using Receiver Operator Characteristic plots.

Results: It proved possible with GPORC to preselect sensitivity and specificity, with an area under the curve of 0.8025. At 95% sensitivity, GPORC had a specificity of 56%, 30% better than audiological protocols with closest sensitivities. Protocols had fixed sensitivities, ranging from 82 to 97%, and specificities between 15% and 61%.

Conclusion: The GPORC model developed increased the flexibility and specificity of the screening process for VS when applied to a large historical sample of matched patients with and without vestibular schwannoma. If applied prospectively and clinically, for instance via an Internet-based freeware, it could reduce the number of “normal” MR scans by as much as 30% without reducing the sensitivity of detecting true cases. Performance of the system can be further improved through incorporating additional data domains into the decision-making process.