Original articles
Neural networks to identify glaucomatous visual field progression

https://doi.org/10.1016/S0002-9394(02)01836-6Get rights and content

Abstract

Purpose

To describe a method to determine progression of glaucoma based on visual field thresholds.

Design

Observational retrospective longitudinal cohort study.

Methods

A back propagation neural network with three hidden layers was developed with commercial software. Visual field data from 80 patients who participated in the Advanced Glaucoma Intervention Study (AGIS) were used. Glaucomatous visual field progression was defined as a change of 4 or more units in the AGIS score, confirmed by at least two sequential subsequent tests. Inputs to the neural network consisted of threshold measurements from 55 visual field locations from the baseline examination and each follow-up examination. The data set was randomized so the sequence of examinations would not influence the training or testing of the neural network. Two thirds of the randomized data were used for training and the remaining one third for testing.

Results

The mean age of 80 patients enrolled in AGIS at initial examination was 67.4 (± 7.3 standard deviation [SD]) years. The average follow-up period was 7.2 (±2.3 SD) years and the mean duration between examinations was 0.46 (± 0.39 SD) years. The neural network estimated the probability of progression for each baseline and follow-up comparison with an average sensitivity of 86% and specificity of 88%. The area under the receiver operating characteristic (ROC) curve was 0.92, with a sensitivity of 86% at the 80% specificity level and a sensitivity of 91% at the 90% specificity level.

Conclusions

From analysis of AGIS data, progression of glaucoma could be detected from visual field thresholds with a neural network.

Section snippets

Design

This study was done as an observational retrospective cohort study. The study took place at The Study Center: Jules Stein Eye Institute, UCLA School of Medicine (Los Angeles, California, USA).

Patients

Data from two AGIS testing centers (Yale University and Georgetown University), collected from October 1988 to January 2000, were used to develop and test the neural network. Study eyes were required to have uncontrolled intraocular pressure (IOP). This means that despite maximally tolerated and effective available treatment, the last two IOP measurements, made on separate days, were both above a critical level, as determined by the AGIS protocol. For an eye with little or no visual field

Results

Data from 80 patients enrolled in the AGIS were used in the current study. The mean age at the time of the initial reference examination was 67.4 ± 7.3 years. The mean follow-up period was 7.2 ± 2.3 years, and the mean duration between adjacent examinations was 0.46 ± 0.39 years (Table 1). Twenty-six patients had both eyes enrolled in AGIS. Analyses for data from these patients were performed in two ways: (1) with one eye randomly selected and (2) with both eyes included.

The following results

Discussion

Physician decision making is often surrounded by a degree of uncertainty, which arises in part from noise in diagnostic tests. A physician must be able to process several pieces of information at once and attempt to distinguish the signal (true diagnosis) from the noise (random fluctuation). This uncertainty is especially large when following a glaucoma patient. In the visual field aspect of glaucoma, there are at least two reasons for the uncertainty: the lack of objectivity in evaluating

References (17)

There are more references available in the full text version of this article.

Cited by (26)

  • Glaucoma progression. Clinical practice guide

    2023, Archivos de la Sociedad Espanola de Oftalmologia
  • A Case for the Use of Artificial Intelligence in Glaucoma Assessment

    2022, Ophthalmology Glaucoma
    Citation Excerpt :

    More recently, the dynamic structure-function model was shown to predict glaucoma progression over a few visits accurately and is generalizable to other testing indices and populations; however, the model is yet to have a clinical functionality.79 The first successful AI implemented to detect progression was completed by Lin et al80 by examining VF to detect progression of glaucoma in patients with an established diagnosis of glaucoma (AUROC of 0.92). Other studies have also implemented VF in AI algorithms for detecting glaucoma.

  • Artificial intelligence and glaucoma: A literature review

    2022, Journal Francais d'Ophtalmologie
  • Digital technology, tele-medicine and artificial intelligence in ophthalmology: A global perspective

    2021, Progress in Retinal and Eye Research
    Citation Excerpt :

    Further studies similarly demonstrated that ML may perform comparably or better than human experts in the mean deviation, pattern standard deviation and glaucoma hemifield test (Goldbaum et al., 2002; Chan et al., 2002). A back-propagation neural network reported to successfully detect visual field progression with an AUC of 0.92 (Lin et al., 2003). Further advances using algorithms based on CNN showed higher sensitivity and specificity than traditional ML methods.

  • Can Big Data change our practices?

    2019, Journal Francais d'Ophtalmologie
  • Neural Network Analysis for the detection of glaucomatous damage

    2014, Applied Soft Computing Journal
    Citation Excerpt :

    Therefore, an alternative way to detect glaucomatous nerve fiber layer damage is to devise an algorithm that can recognize the changes from normal pattern. The application of artificial neural network has been reported in classification and interpretation of visual field [9], assessment of visual field progression [10] and identification of glaucomatous damage using optic nerve and visual field variables [11]. In assessing the performance of a detection method, the Receiver Operating Characteristic (ROC) is often used for binary classifiers.

View all citing articles on Scopus
View full text