Pattern recognition of vertical strabismus using an artificial neural network (StrabNet)

BACKGROUND: Our goal is to develop an accessible expert system (StrabNet) that will assist in the clinical diagnosis of vertical strabismus, form the basis of a teaching/learning tool, and contribute to the audit process. Potentially, this model can be extended to other strabismus deviations. METHODS: Vertical deviations were separated into eight classifications (diagnoses). An expert system based on architecture of artificial neural networks learned the patterns for each class of deviation based on 10 prism cover-test measurements (9 cardinal positions and near fixation). The expert system was tested with previously unseen and real-patient data. This system was extended to a reduced model requiring only six measurements (primary position, right, left, up, down gaze, and near fixation), and evaluated with real patient data. A freely available Web implementation is available on the Internet at www.StrabNet.com . RESULTS: The expert system was found to be highly accurate at diagnosing vertical strabismus (100% and approximately 96% for ten and six measurement models, respectively) from one of the eight classes. CONCLUSION: StrabNet is of demonstrable value in diagnosing commonly presenting vertical deviations from prism cover test (PCT) measurements. Its potential role in teaching and in audit is identified.