Purchase this article with an account.
J. Garcia–Sanchez, J. Garcia–Feijoo, E. Carmona, A.M. Fernandez–Vidal, C.D. Mendez–Hernandez, A. Castillo–Gomez, J.M. Benitez del Castillo, J. Mira; Application of a neurofuzzy system to the automatic interpretation of the visual field in glaucoma. . Invest. Ophthalmol. Vis. Sci. 2004;45(13):3304.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: To study the possible clinical applications of a neurofuzzy system (NEFCLASS) to the automatic interpretation of the visual fields. Methods: 218 eyes of 218 primary open angle glaucoma patients (Mean age: 58,7 SD 14,7) and 62 eyes of 62 control subjects (55,4 SD 15,8) were analyzed. Inclusion criteria:VA≥ 0.5, IOP < 20 mmHg (under treatment). Refraction ± 3.00 D, previous perimetric experience, stable VF defects (6 months). Exclusion criteria: Miotics, other ocular pathology, diabetes, hypertension, neurological illness. Patients were recruited to obtain the following VF defect distribution: 50% with early VF defect, 30% moderate and 20% advanced. Visual field analysis was performed using the Octopus G1 program (OCTOPUS 123, Interzeag AG, Switzerland). We considered 7 VF areas. In each area MD and LV were calculated. We also calculated the number of points with a defect > 5 dB. VF were classified and separated in a training sample (2/3 of the VF for each group: 186) and a test sample (the remaining 1/3: 32). The neurofuzzy system used was the Neuro–Fuzzy CLASSIFIER (by Nauck and Kruse; Magdeburg, Germany). This system enables fuzzy rules to be obtained from data for classifying patterns in a specific number of classes. It uses a supervised learning algorithm based on fuzzy error back propagation. Results: We defined a set of 6 neurofuzzy rules for the classification of visual fields. Then, different trials were carried out to determine the number of partitions of the variables selected. We report the results of the final trial: a total of 8 visual fields were misclassified by the network (4 normals and 4 glaucomatous VFs). 6 mistakes correspond to the training sample (4N and 2G) and 2 to the test sample (0N and 2G). Global sensitivity was 98,2% and global specificity was 93,5%. Conclusions: This neurofuzzy system seems to be a good approach to the objective classification of visual fields in glaucoma. Good classification percentages were obtained.
This PDF is available to Subscribers Only