PURPOSE: To develop a neural network model for planning of the surgical strategy of patients with sensorial strabismus. METHODS: In this retrospective study, medical records of 95 patients with sensorial strabismus were reviewed. All patients were seen at the Strabismus Sector of the Hospital das Clínicas of the University of São Paulo. The neural network was designed containing 3 layers. Sixty-eight patients were used in the training and validation set, and 27 in the test set. RESULTS: In the 68 patients used in the training and validation set, 37 had exotropia, and 31 esotropia. The backpropagation approach was used for training the neural network. A learning rate of 0.6, and a tolerance error of 0.05 were used. In the 27 patients used in the test set, 18 had exotropia, and 9 had esotropia. The efficacy of the neural network was analyzed using the average of the difference between the indication supplied by the network and the original indication. In patients with exotropia, the average error was 0.4 mm (±0.4), for recession of the lateral rectus muscle, and 0.3 mm (±0.3), for the resection of the medial rectus muscle. In the esotropia group, the average error was 0.2 mm (±0.2) for the recession of the medial rectus muscle, and 0.5 mm (±0.3) for resection of the lateral rectus muscle. CONCLUSION: As the artificial neural network can simulate a biological central nervous system, and is able to carry out cognitive tasks, it can be a viable option to help the surgical planning for strabismus correction.
Strabismus; Surgical procedures, operative; Esotropia; Exotropia; Nerve Net
