Predicting visual field global and local parameters from OCT measurements using explainable machine learning

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Фев. 16, 2025

Abstract Glaucoma is characterised by progressive vision loss due to retinal ganglion cell deterioration, leading gradual visual field (VF) impairment. The standard VF test may be impractical in some cases, where optical coherence tomography (OCT) can offer predictive insights into for multimodal diagnoses. However, predicting measures from OCT data remains challenging. To address this, five regression models were developed predict OCT, Shapley Additive exPlanations (SHAP) analysis was performed interpretability, and a clinical software tool called Predictor developed. evaluate the models, total of 268 glaucomatous eyes (86 early, 72 moderate, 110 advanced) 226 normal included. machine learning outperformed recent OCT-based prediction deep studies, with correlation coefficients 0.76, 0.80 0.76 mean deviation, index pattern respectively. Introducing pointwise normalisation step-size concept, absolute error 2.51 dB obtained sensitivity prediction, grayscale model yielded structural similarity 77%. SHAP-based provided critical most relevant features glaucoma diagnosis, showing promise assisting eye care practitioners through an explainable AI tool.

Язык: Английский

---

OCT-based diagnosis of glaucoma and glaucoma stages using explainable machine learning

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Янв. 28, 2025

Abstract Glaucoma poses a growing health challenge projected to escalate in the coming decades. However, current automated diagnostic approaches on diagnosis solely rely black-box deep learning models, lacking explainability and trustworthiness. To address issue, this study uses optical coherence tomography (OCT) images develop an explainable artificial intelligence (XAI) tool for diagnosing staging glaucoma, with focus its clinical applicability. A total of 334 normal 268 glaucomatous eyes (86 early, 72 moderate, 110 advanced) were included, signal processing theory was employed, model interpretability rigorously evaluated. Leveraging SHapley Additive exPlanations (SHAP)-based global feature ranking partial dependency analysis (PDA) estimated decision boundary cut-offs machine (ML) novel algorithm developed implement XAI tool. Using selected features, ML models produce AUC 0.96 (95% CI: 0.95–0.98), 0.98 0.96–1.00) 1.00 1.00–1.00) respectively differentiating moderate advanced glaucoma patients. Overall, outperformed clinicians early stage overall 10.4 –11.2% higher accuracy. The user-friendly software shows potential as valuable eye care practitioners, offering transparent interpretable insights improve decision-making.

Язык: Английский

---

Deep Learning Approaches Predict Glaucomatous Visual Field Damage from OCT Optic Nerve Head En Face Images and Retinal Nerve Fiber Layer Thickness Maps

Ophthalmology, Год журнала: 2019, Номер 127(3), С. 346 - 356

Опубликована: Сен. 30, 2019

Язык: Английский

---

Clinical Evaluation of Swedish Interactive Thresholding Algorithm–Faster Compared With Swedish Interactive Thresholding Algorithm–Standard in Normal Subjects, Glaucoma Suspects, and Patients With Glaucoma

American Journal of Ophthalmology, Год журнала: 2019, Номер 208, С. 251 - 264

Опубликована: Авг. 27, 2019

To compare the visual fields results obtained using Swedish interactive thresholding algorithm-Standard (SS) and algorithm-Faster (SFR) in normal subjects, glaucoma suspects, patients with to quantify potential time-saving benefits of SFR algorithm. Prospective, cross-sectional study. One randomly selected eye from 364 (77 178 109 glaucoma) seen a single institution underwent testing both SS on Humphrey Field Analyzer. Cumulative test time each algorithm was compared after accounting for different rates reliability. Pointwise cluster analysis performed determine whether there were systematic differences between algorithms. Using had greater rate unreliable (29.3%) (7.7%, P < .0001). This mainly because high false positive seeding point errors. However, modeled times showed that could obtain number reliable within shorter period time. resulted higher sensitivity values (on average 0.5 dB under conditions field loss (<19 dB). Cluster no patterns After reliability, can result significant savings SS. Clinicians should be cognizant errors as common sources error SFR. Results algorithms are not directly interchangeable, especially if is deficit <19 dB.

Язык: Английский

---

Ability of 24-2C and 24-2 Grids to Identify Central Visual Field Defects and Structure-Function Concordance in Glaucoma and Suspects

American Journal of Ophthalmology, Год журнала: 2020, Номер 219, С. 317 - 331

Опубликована: Июль 1, 2020

•24-2C returns global results similar to those of 24-2. Use 24-2C in glaucoma staging is unclear.•24-2C may be used a targeted approach where central defects are suspected.•24-2C not ideal for comparing loss using the ganglion cell analysis glaucoma. PurposeThe purpose this study was compare ability 24-2 test grid with that measure visual field indices, identify defects, and facilitate macular structure-function optical coherence tomography (OCT) scans suspects patients.DesignProspective, cross-sectional study.MethodsOne eye from each 100 patients (60 undergoing SITA-Faster [Zeiss Meditec] testing on 24-2C; 40 SITA-Standard 24-2C) were included study. Global duration, pattern deviation extracted. The map Cirrus OCT (Carl Zeiss Meditec) Ganglion Cell Analysis (GCA) extracted, relationships compared after correction stimulus location stimulated corresponding retinal cell.ResultsGlobal index grid, both identified comparable number clusters defects. Centrally, more than but statistically significant. Although locations resulted instances concordance locations, half fell close or outside GCA when corrected displacement.ConclusionsThe returned indices tended functional Central better achieved did coincide commonly thickness scan. patients. Prospective, One cell. displacement.

Язык: Английский

---

Clinical outcomes of the Centre for Eye Health: an intra-professional optometry-led collaborative eye care clinic in Australia

Clinical and Experimental Optometry, Год журнала: 2021, Номер 104(7), С. 795 - 804

Опубликована: Фев. 25, 2021

Henrietta Wangab & Michael Kalloniatisab* a Centre for Eye Health, Sydney, Australiab School of Optometry and Vision Science, University New South Wales, Australia

Язык: Английский

---

Comparison of 10-2 and 24-2C Test Grids for Identifying Central Visual Field Defects in Glaucoma and Suspect Patients

Ophthalmology, Год журнала: 2021, Номер 128(10), С. 1405 - 1416

Опубликована: Март 18, 2021

PurposeTo compare the ability of 24-2C and 10-2 test grids in measuring visual field global indices, identifying central defects, facilitating macular structure-function analysis with OCT scans glaucoma suspect patients.DesignProspective, cross-sectional study.ParticipantsOne eye from 131 57 patients recruited a referral-only, university-based clinic.MethodsEach subject underwent perimetric testing using SITA-Faster SITA-Fast random order, Cirrus imaging (Ganglion Cell Analysis) for correlations.Main Outcome MeasuresVisual indices (mean deviation, pattern standard binarized "cluster" pass/fail, mean sensitivity), number proportion concordance deviation map following location displacement correspondence underlying retinal ganglion cell position.ResultsGlobal sensitivity) were similar between both grids. The detected more defects compared (P < 0.0001 all patients, P = 0.006 patients). This was preserved when analyzing 0.02) but not significantly different 0.051). identified "clusters" 2+ contiguous points deficit 0.0001). Structure-function comparisons performed at locations where colocalized revealed greater structural functional deficits took median 201 seconds, 154 corresponding to thresholding algorithms.ConclusionsThe return performance proportionally amounts loss. additional grid rate grid. Thus, can identify presence clustered defect probability criteria, whereas may be useful comprehensively characterizing predicting function. To patients. Prospective, study. One clinic. Each correlations. Visual position. Global algorithms.

Язык: Английский

---

Development of a Spatial Model of Age-Related Change in the Macular Ganglion Cell Layer to Predict Function From Structural Changes

American Journal of Ophthalmology, Год журнала: 2019, Номер 208, С. 166 - 177

Опубликована: Май 10, 2019

PurposeTo develop location-specific models of normal, age-related changes in the macular ganglion cell layer (GCL) from optical coherence tomography (OCT). Using these OCT-derived models, we predicted visual field (VF) sensitivities and compared results to actual VF sensitivities.DesignRetrospective cohort study.MethodsSingle eyes 254 normal participants were retrospectively enrolled Centre for Eye Health (Sydney, Australia). Macular GCL measurements obtained using Spectralis OCT. Cluster algorithms performed identify spatial patterns demonstrating similar change. Quadratic linear regression subsequently used characterize decline. Forty underwent additional testing with Humphrey VFs, 95% prediction intervals calculated measure predictive ability structure-function incorporating cluster-based pooling, age correction, consideration summation.ResultsQuadratic provided a superior fit (P value <.0001–.0066), establishing that decline commences late 30s across macula. The equivalent rates showed eccentricity-dependent variation (0.13 μm/yr centrally vs 0.06 peripherally); however, average, normalized loss per year was consistent 64 measurement locations at 0.26%. describing significantly narrower all (3.79–4.99 dB) without clustering applied (5.66–6.73 dB, P < .0001).ConclusionsCombining age-correction based on allowed development robust age. resultant sensitivity may be future disease improve detection early pathology. To sensitivities. Retrospective study. Single summation. .0001). Combining

Язык: Английский

---

Consistency of Structure-Function Correlation Between Spatially Scaled Visual Field Stimuli and In Vivo OCT Ganglion Cell Counts

Investigative Ophthalmology & Visual Science, Год журнала: 2018, Номер 59(5), С. 1693 - 1693

Опубликована: Апрель 2, 2018

Purpose: To investigate the effect of stimulus size and disease status on structure-function relationship within central retina, we correlated differential light sensitivity (DLS) with Goldmann I to V (GI–V) optical coherence tomography (OCT) derived in vivo ganglion cell count per area (GCc) macular normal subjects patients early glaucoma. Methods: Humphrey Field Analyzer 10–2 visual field data GI through Spectralis OCT layer (GCL) thickness measurements were collected from glaucoma cohorts including 25 each. GCc was calculated GCL DLSs for different sizes. Results: Correlation coefficients attained smaller higher compared larger sizes both (GI–GII: R2 = 0.41–0.43, GIII–GV: 0.16–0.41) diseased 0.33–0.41, 0.19–0.36). Quadratic regression curves combined demonstrated high correlation (R2= 0.82–0.90) differed less than 1 dB range between cohorts. The established compatible a histologically model spanning predicted by GIII. Conclusions: Stimulus critical spatial summation (GI–II) improved correlations field. identical cohort when GV combined. Congruency GII those previously GIII more peripheral locations further suggests that is governed number area.

Язык: Английский

---

An evidence‐based approach to the routine use of optical coherence tomography

Clinical and Experimental Optometry, Год журнала: 2018, Номер 102(3), С. 242 - 259

Опубликована: Дек. 18, 2018

Optical coherence tomography is an imaging technology that has revolutionised the detection, assessment and management of ocular disease. It now a mainstream in clinical practice performed by non‐specialised personnel some settings. This article provides perspective on implications movement describes best using multimodal evidence‐based approach. Practical, illustrative guides interpretation optical are provided for three major diseases fundus, which often crucial to management: age‐related macular degeneration, diabetic retinopathy glaucoma. Topics discussed include: cross‐sectional longitudinal signs disease, so‐called 'red‐green' disease whereby clinicians rely machine/statistical comparisons diagnosis managing treatment‐naïve patients, utility angiography machine learning.

Язык: Английский

---