Global Search

Search articles, concepts, and chapters

Transl Vis Sci TechnolOctober 20251 citations

A Deep Learning Model for Glaucoma Detection Outperforms Conventional Metrics.

Lau Wai Tak, Tsamis Emmanouil, Hood Donald C, Thakoor Kaveri


AI Summary

A deep learning model for glaucoma detection from OCT reports showed superior sensitivity and specificity compared to conventional metrics, offering a promising tool for screening and clinical diagnosis.

Abstract

Purpose

To test a deep learning model (DLM) for detecting glaucomatous damage with optical coherence tomography (OCT) reports.

Methods

An ImageNet-pretrained ResNet-50 model was fine-tuned to classify healthy (H) versus glaucomatous (G) OCT reports. The model was trained with 4932 H and 207 G OCT wide-field reports and tested on three unseen datasets. The "Clear G" cohort consisted of 50 eyes with clear glaucomatous damage based on OCT, visual fields, photos, and clinical exams. The "Pattern ON-G" cohort contained 183 eyes over 60 years of age with OCT arcuate defects. These defects ranged from subtle to advanced. The H-test cohort contained 396 eyes from a commercial normative reference database. We compared the DLM to five conventional metrics.

Results

For a DLM cutoff yielding a 99.5% specificity on the 396 H-test eyes, the DLM achieved a sensitivity of 100% with the 50 Clear G and 95.1% with the 183 Pattern ON-G reports. The AUROC was 0.999. The DLM was significantly better than 3 conventional metrics. For example, global circumpapillary retinal nerve fiber thickness showed a specificity of 97.7% for the 396 H eyes, sensitivities of 73.8% and 44.1% for the G cohorts, and an area under the receiver operating characteristic curve (AUROC) of 0.883. The best of the conventional metrics, which was based on a logistic regression model, also had significantly poorer sensitivities (84.6% and 66.1%) for the G cohorts and a lower AUROC (0.975).

Conclusions

The DLM achieved high specificity and sensitivity and outperformed conventional metrics.

Translational relevance: The DLM provides a possible method for screening, and a possible diagnostic aid for clinicians.


MeSH Terms

HumansDeep LearningGlaucomaTomography, Optical CoherenceMiddle AgedMaleFemaleVisual FieldsAgedROC CurveSensitivity and SpecificityRetinal Ganglion Cells

Key Concepts5

A deep learning model (DLM) for detecting glaucomatous damage with optical coherence tomography (OCT) reports, fine-tuned from an ImageNet-pretrained ResNet-50 model, achieved a sensitivity of 100% with 50 eyes in the 'Clear G' cohort (clear glaucomatous damage based on OCT, visual fields, photos, and clinical exams) at a cutoff yielding 99.5% specificity on 396 H-test eyes.

DiagnosisCross-sectionalModel Testingn=50 eyes (Clear G cohort), 396 eyes (H…Ch5Ch6Ch12

A deep learning model (DLM) for detecting glaucomatous damage with optical coherence tomography (OCT) reports, fine-tuned from an ImageNet-pretrained ResNet-50 model, achieved a sensitivity of 95.1% with 183 eyes in the 'Pattern ON-G' cohort (over 60 years of age with OCT arcuate defects) at a cutoff yielding 99.5% specificity on 396 H-test eyes.

DiagnosisCross-sectionalModel Testingn=183 eyes (Pattern ON-G cohort), 396 e…Ch5Ch6Ch12

The deep learning model (DLM) for detecting glaucomatous damage with optical coherence tomography (OCT) reports achieved an AUROC of 0.999, which was significantly better than 3 conventional metrics.

Comparative EffectivenessCross-sectionalModel Comparisonn=50 Clear G eyes, 183 Pattern ON-G eye…Ch5Ch6Ch12

Global circumpapillary retinal nerve fiber thickness, as a conventional metric for detecting glaucomatous damage, showed a specificity of 97.7% for 396 H eyes, sensitivities of 73.8% (50 Clear G eyes) and 44.1% (183 Pattern ON-G eyes) for the glaucomatous cohorts, and an AUROC of 0.883.

DiagnosisCross-sectionalModel Comparisonn=50 Clear G eyes, 183 Pattern ON-G eye…Ch5Ch6Ch12

The best conventional metric, based on a logistic regression model for detecting glaucomatous damage, had sensitivities of 84.6% (50 Clear G eyes) and 66.1% (183 Pattern ON-G eyes) for the glaucomatous cohorts and an AUROC of 0.975, which was significantly poorer than the deep learning model.

Comparative EffectivenessCross-sectionalModel Comparisonn=50 Clear G eyes, 183 Pattern ON-G eye…Ch5Ch6Ch12

Is this article assigned to the wrong chapter(s)? Let us know.