Using unsupervised learning with independent component analysis to identify patterns of glaucomatous visual field defects

PURPOSE: Clustering by unsupervised learning with machine learning classifiers was shown to segment clusters of patterns in standard automated perimetry (SAP) for glaucoma in previous publications. In this study, unsupervised learning by independent component analysis decomposed SAP field patterns into axes, and the information represented by these axes was evaluated. METHODS: SAP fields were used that were obtained with the Humphrey Visual Field Analyzer (Carl Zeiss Meditec, Dublin, CA) from 189 normal eyes and 156 eyes with glaucomatous optic neuropathy (GON) determined by masked review with stereoscopic optic disc photographs. The variational Bayesian independent component analysis mixture model (vB-ICA-mm) partitioned the SAP fields into the most informative number of clusters. Simultaneously, the model learned an optimal number of maximally independent axes for each cluster. RESULTS: The most informative number of clusters in the SAP set was two. vB-ICA-mm placed 68.6% of the eyes with GON in a cluster labeled G and 98.4% of the eyes with normal optic discs in a cluster labeled N. Cluster G optimally contained six axes. Post hoc analysis of patterns generated at -1 SD and +2 SD from the cluster G mean on the six axes revealed defects similar to those identified by experts as indicative of glaucoma. SAP fields associated with an axis showed increasing severity, as they were located farther in the positive direction from the cluster G mean. CONCLUSIONS: vB-ICA-mm represented the SAP fields with patterns that were meaningful for glaucoma experts. This process also captured severity in the patterns uncovered. These findings should validate vB-ICA-mm as a data-mining technique for new and unfamiliar complex tests.     

Effectiveness of unsupervised oculokinetic perimetry for detecting glaucomatous visual field defects

Oculokinetic perimetry (OKP) is a simple, inexpensive technique which was introduced primarily to promote more widespread use of visual field screening for glaucoma. The effectiveness of unsupervised OKP screening was evaluated using 145 non-glaucomatous patients and 64 patients with previously undiagnosed primary open angle glaucoma. Glaucoma cases were validated by the extended l32 point program on the Henson CFS2000 instrument. Both sample populations were representative of patients presenting for primary vision care. The effectiveness of the 26 point OKP test was compared with the 26 point Henson screening program. The results suggest that, in contrast to the Henson screening test (sensitivity = 85.1 %. specificity = 93.8%). unsupervised OKP screening (sensitivity = 25.0%, specificity = 93.6%) has limited effectiveness for detecting glaucomatous visual field defects.     

Neural networks to identify glaucomatous visual field progression

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.     

Using machine learning classifiers to identify glaucomatous change earlier in standard visual fields

PURPOSE: To compare the ability of several machine learning classifiers to predict development of abnormal fields at follow-up in ocular hypertensive (OHT) eyes that had normal visual fields in baseline examination. METHODS: The visual fields of 114 eyes of 114 patients with OHT with four or more visual field tests with standard automated perimetry over three or more years and for whom stereophotographs were available were assessed. The mean (+/-SD) number of visual field tests was 7.89 +/- 3.04. The mean number of years covered (+/-SD) was 5.92 +/- 2.34 (range, 2.81-11.77). Fields were classified as normal or abnormal based on Statpac-like methods (Humphrey Instruments, Dublin, CA) and by several machine learning classifiers. The machine learning classifiers were two types of support vector machine (SVM), a mixture of Gaussian (MoG) classifier, a constrained MoG, and a mixture of generalized Gaussian (MGG). Specificity was set to 96% for all classifiers, using data from 94 normal eyes evaluated longitudinally. Specificity cutoffs required confirmation of abnormality. RESULTS: Thirty-two percent (36/114) of the eyes converted to abnormal fields during follow-up based on the Statpac-like methods. All 36 were identified by at least one machine classifier. In nearly all cases, the machine learning classifiers predicted the confirmed abnormality, on average, 3.92 +/- 0.55 years earlier than traditional Statpac-like methods. CONCLUSIONS: Machine learning classifiers can learn complex patterns and trends in data and adapt to create a decision surface without the constraints imposed by statistical classifiers. This adaptation allowed the machine learning classifiers to identify abnormality in visual field converts much earlier than the traditional methods.     

Classifying patterns of localized glaucomatous visual field defects on automated perimetry

AIM: To classify the classic patterns of glaucomatous visual field defects on automated perimetry and to study their proximity to fixation. STUDY DESIGN: Cross-sectional observational study. MATERIALS AND METHODS: About 1120 full threshold 30-2 reliable visual fields of glaucoma patients were analyzed by 2 glaucomatologists. Classically described patterns of visual field defects were identified on the pattern deviation plot and definitions proposed. Interreader agreement between 3 independent (not involved in the classification) readers was determined. Proximity to fixation of the different patterns was assessed. RESULTS: Interreader agreement with 3 readers was found to be 93% or more between any 2 readers using the present system of classification. Central fixation was seen to be involved in 45% of the glaucomatous visual field defects studied overall. CONCLUSIONS: The proposed definitions of topographical glaucomatous field defects based on the pattern deviation probability plot are simple to use in clinical practice with good interreader agreement.     

Progression of early glaucomatous visual field defects

We studied the mode of progression of visual field defects in glaucoma the using static computerized perimetry (OCTOPUS 201). As a result of analysis using Delta Program, the modes of progression of the visual field defects were divided into two groups, general depression and local depression. The cases of general depression were younger than those of local depression. In cases of local depression, we examined the direction of progression in isolated scotomas. Isolated scotomas located at the nasal visual field tended to progress toward the nasal and peripheral side.     

Color perimetry of glaucomatous visual field defects

A color video tangent screen has been devised, using microcomputer control of a video display to produce colored perimetric test objects matched in luminance to a white surround at 10-foot lamberts . Perimetric isopters for varying degrees of color saturation were determined by kinetic perimetry. This form of color perimetry was used to examine one eye of each of 40 patients with open-angle glaucoma as well as 20 glaucoma-suspect patients. For the first 23 eyes with manifest glaucomatous visual field defects, a masked comparison was made between the results of color perimetry and conventional perimetry with a Goldmann perimeter. For these 23 eyes, color perimetry did as well as luminance perimetry in 14, was less sensitive in 2, and was more sensitive in 7. All defects that were detectable by conventional perimetry were successfully demonstrated by the color method. Such defects often appeared to be greater in extent when mapped by the color method as compared to conventional luminance perimetry.     

Quantification of glaucomatous visual field defects with automated perimetry

A method to quantify different glaucomatous visual field defects is presented. Three visual field indices are calculated: the short-term fluctuation, the mean defect, and the corrected loss variation. The method was applied to visual fields tested with program JO on the Octopus automated perimeter. The indices of 130 glaucoma suspects and 50 glaucoma patients were compared with 100 normal controls. The indices provide good detectability of visual field defects and easy follow-up.     

Unsupervised machine learning with independent component analysis to identify areas of progression in glaucomatous visual fields

PURPOSE: To determine whether a variational Bayesian independent component analysis mixture model (vB-ICA-mm), a form of unsupervised machine learning, can be used to identify and quantify areas of progression in standard automated perimetry fields. METHODS: In an earlier study, it was shown that a model using vB-ICA-mm can separate normal fields from fields with six different patterns of visual field loss related to glaucomatous optic neuropathy (GON) along maximally independent axes. In the present study, an independent group of 191 patient eyes (66 with ocular hypertension (OHT), 12 with suspected glaucoma by field, 61 with suspected glaucoma by disc, and 52 with glaucoma) with five or more standard visual fields under observation for a mean of 6.24 +/- 2.65 years and 8.11 +/- 2.42 visual fields were evaluated with the vB-ICA-mm. In addition, eyes with progressive GON (PGON) were identified (n = 39). Each participant had a series of fields tested, with each field entered independently and placed along the axes of the previously developed model. This allowed change in one pattern of visual field defect (along one axis) to be assessed relative to results other areas of that same field (no change along other axes). Progression was based on a slope falling outside the 5th and the 95th percentile limits of all slopes, with at least two axes not showing such a deviation in a given individual's series of fields. Fields were also scored using Advanced Glaucoma Intervention Study (AGIS) and the Early Manifest Glaucoma Treatment Trial (EMGT) criteria. RESULTS: Thirty-two of 191 eyes progressed on vB-ICA-mm by this definition. Of the 32, 22 had field loss at baseline, 7 had only GON, 3 were OHTs and 12 were from the 39 eyes (31%) with PGON. The vB-ICA-mm identified a higher percentage of progressing eyes in each diagnostic category than did AGIS or and the EMGT. CONCLUSIONS: The vB-ICA-mm can quantitatively identify progression in eyes with glaucoma by evaluating change in one or more patterns of the visual field loss while other areas or patterns remain stable. This may enable each eye to contribute to the determination of whether change is caused by true progression or by variability.     

Frequency distribution of early glaucomatous visual field defects

The optimization of visual field screening programs for glaucoma requires precise knowledge of where glaucoma defects occur and whether any relation exists between the size of defects and the associated retinal disease. The visual field results from 109 eyes of patients with early glaucomatous damage were investigated with a Friedmann Mark II Visual Field Analyser. The frequency distribution of scotomas as a function of location demonstrated that the tested field could be divided into four zones, one where scotomas occur frequently, one where scotomas occur a reasonable number of times, one where scotomas occur rarely, and one which corresponds to the blind spot (BS) region. The distribution of scotoma size in the first three zones was found to be about the same. Differences among the distributions of defects measured in this study and those of previous studies are attributed to different examination techniques.     

Logistic regression analysis of risk factors of glaucomatous visual field defects]

30 patients (42 eyes) of early primary glaucoma with normal visual field were followed up for 2.5 years, during which time visual field defects developed in 24 eyes. Multivariate logistic regression analysis of the initial values of suspected risk factors demonstrated that a positive family history, the vertical C/D ratio, and the diurnal peak IOP were significant in predicting impending visual field defects.     

Reversibility of glaucomatous defects of the central visual field

We analyzed more than 4 consecutive measurements of the central area of the visual field in 1,004 eyes of 533 subjects with various types of glaucoma as diagnosed by the Octopus 201 and its examination program No. 31. We used a non-repeated analysis of variance to determine the statistical relationship between the mean sensitivity measurements for individual quadrants of the central visual field. We made the series of measurements used in this statistical analysis during repeated examinations that took place over a follow-up period of 1 year. Out of 1,004 eyes, 53 eyes (5.3%) were shown to have an improved mean sensitivity of the central 30 degrees of the visual field. We adopted the degree of sensitivity improvement [(final mean sensitivity - initial mean sensitivity)/follow-up period] as the indicator of the degree of reversibility in the central visual field to determine which factors might influence changes in the visual field. In patients under 50 years of age, who were at an early stage of the disease, as opposed to those over 50 years of age, who were at an intermediate or advanced stage of this disease or who had undergone reduction of intraocular pressure, a relatively high degree of sensitivity improvement was more frequently observed. In the cases with primary open angle glaucoma which includes low tension glaucoma, we think treatment for abnormality in the aqueous humor dynamics is very important. We concluded, that the reversibility of the central visual field defect in glaucomatous eyes is possible, though it can only be clearly detected by careful follow-up examinations employing computerized perimetry.     

Variability of glaucomatous visual field defects in computerized perimetry

Summary Normal subjects are able to maintain a stable level of performance throughout a test session of 30 min, using 0.5 s exposure time. Test points placed in normal areas of glaucoma cases behave similarly, whereas test points in relative scotomatous parts of the field show not only a lower sensitivity level but also a tendency to a decline of the performance level and/or a higher threshold variation.Change of exposure time showed remarkable effects on the performance level at defect points, an increase to 1 s causing a stabilization on an improved level and a decrease to 0.25 s giving a deterioration.The results are discussed in terms of noise admixture to the signal. According to this model an increased noise amplitude by impaired retinal function reduces the performance level and augments the variation. On the other hand, an increased exposure time of the test object, which causes a stabilization of the process on an improved performance level, may be interpreted as a reduction of the noise amplitude.

---

Zusammenfassung Gesunde Probanden zeigen bei der Perimetrie bis zu einer Untersuchungszeit von 30 min nur unbedeutliche Änderung der Schwellenwerte der Netzhautempfindlichkeit bei einer Darbietungszeit der Stimuli von 0.5 s Testpunkte im gesunden Gesichtsfeldbereich von Glaukompatienten verhalten sich ähnlich, während Testpunkte im Bereich relativer Gesichtsfeldausfälle nicht nur geringere sondern auch eine abnehmende Empfindlichkeit und/oder eine größere Variabilität der Schwellenwerte aufweisen.Eine Veränderung der Darbietungszeit zeigte einen erheblichen Effekt auf den Schwellenwerten bei relativen Skotomen; eine Zunahme der Darbietungszeit auf 1 s zeigte eine Stabilisierung auf erhöhter Empfindlichkeit, eine Abnahme der Darbietungszeit auf 0.25 s ergab eine starke Verschlechterung.Die Ergebnisse werden diskutiert unter dem Aspekt der Informationsübertragung unter Beimengung von Rauschen. Nach diesem Modell könnte eine vergrößerte Amplitude des Rauschens bei herabgesetzter Netzhautfunktion eine Verschlechterung des Schwellenwertes und die größere Variabilität bedingen. Anderseits kann die verlängerte Expositionszeit mit einer Stabilisierung des verbesserten Schwellenwertes als eine Reduzierung der Rauschen-Amplitude gewertet werden.

     

Mapping of glaucomatous visual field defects by multifocal VEPs.

PURPOSE: To objectively measure a visual field and to sensitively detect glaucomatous visual field defects by measuring the contrast sensitivity function (CSF), using multifocal visual evoked cortical potentials (MVEPs). METHODS: MVEPs of normal subjects (n = 28) and of patients with glaucoma (n = 12) or ocular hypertension (OH, n = 1) were recorded. A multi-input procedure was used to obtain 37 local VEP responses to each scaled hexagon, composed of 24 triangular patterns, reversing in a counterphase manner. Two pattern contrasts of 32% and 8% were used for measuring the contrast threshold. To improve the signal-to-noise ratio, 37 MVEPs were averaged into 20 groups. The root-mean-square (RMS) measures at both contrasts were calculated. Contrast thresholds were estimated by extrapolating the regression line of the amplitude versus contrast to the mean noise levels. RESULTS: RMS amplitudes of each local MVEP decreased as the eccentricity increased and as the pattern contrast decreased in normal subjects. It was also revealed that the amplitudes were smaller in the upper-half field than those in the lower-half field. Compared with the RMS amplitudes, CSFs estimated by MVEPs were relatively constant without being strongly influenced by retinal eccentricity. In patients with glaucoma, the CSFs, even from the locations where the mean perimetric sensitivities ranged to more than 30 dB, were significantly smaller than those in the normal control subjects (P < 0.001, Mann-Whitney test). CSFs in the 20- to 30-dB and 100 to 20-dB groups were also significantly smaller than those in the more-than-30-dB or 20- to 30-dB groups, respectively (P < 0.001, Mann-Whitney test). A significant correlation was found between the logarithmic function of the CSF and perimetric sensitivity (r = 0.57, P < 0.001, n = 216). The CSFs were evaluated on the basis of normal SD. Mappings of the CSFs agreed well with those of the perimetric sensitivity in all patients with glaucoma and was even more sensitive in detecting slight optic nerve damage by glaucoma than was perimetry. CONCLUSIONS: Local optic nerve damage caused by glaucoma can be sensitively detected by measuring contrast sensitivity using the MVEP.     

多焦视诱发电位在青光眼视野损害检测中的应用

目的了解多焦视诱发电位(mfVEP)在青光眼视野损害检测中的作用。设计病例对照研究。研究对象 20例正常人和20例原发性青光眼患者。方法对20例正常人和20例原发性青光眼患者应用RETIscan mfVEP记录系统3．20行检查, 分别测量所有受试者mfVEP每一位点记录曲线的振幅,并计算受试者的双眼视网膜反应不对称系数(RAC)。比较青光眼患者 mfVEP检查的暗点与OCTOPUS 101型自动视野计G2程序检查的视野缺损区的对应性。主要指标振幅、RAC。结果在与自动视野计检查发现的视野缺损相对应的部位,90％患眼的mfVEP表现为至少3个相邻位点的振幅降低或RAC超出正常人数据库均数的2倍标准差(SD)。其中63％患眼表现为至少3个相邻位点的振幅降低超出正常人均数的2 SD,70％患眼表现为至少3个相邻位点的RAC超出正常人均数的2 SD。有2例患者(2眼)的自动视野计检查结果正常,但mfVEP显示有视野异常。正常人中无一眼的mfVEP表现为视野异常。结论 mfVEP有可能作为青光眼视野损害的客观检查手段。     

Detection of progressive glaucomatous papilla changes before onset of visual field defects

Computerized disc analysis was used for the longitudinal monitoring of patients with moderate intraocular pressure elevation. 44 patients (80 eyes) were monitored for at least 18 months; 25 patients (47 eyes) were monitored for at least 12 months; 15 patients (25 eyes) were monitored for at least 18 months. On average, one computerized disc analysis was performed every 4 months. Eight eyes showed a progressive decrease in their neuroretinal rim area. Seven of them were monitored for more than 18 months, one was monitored for 15 months. In all cases, the progressive decrease in neuroretinal rim area was significant, p less than or equal to .1. It is non-physiological and has to be interpreted as a serious sign of ongoing glaucoma damage. In addition, it obviously precedes the onset of visual field defects. Seven out of 8 eyes with progressive decrease in their neuroretinal rim area have normal visual fields up to now. In one case a progressive decrease in neuroretinal rim area was found at first and small visual field defects developed subsequently.