Concepts

Glossary

Shorter terms used across the pipeline, the patient cards and these pages.

OD / OS / OU

Eye & clinic

Latin shorthand used on eye records: OD is the right eye (oculus dexter), OS the left eye (oculus sinister), OU both eyes.

Retinal ganglion cells

Eye & clinic

The retina's output neurons. Their axons run across the retina, gather at the optic disc and form the optic nerve. Glaucoma is the gradual loss of these cells.

Neuroretinal rim

Eye & clinic

The ring of nerve tissue between the edge of the cup and the edge of the disc. It holds the axons, so it thins as glaucoma advances.[9]

ISNT rule

Eye & clinic

In most healthy discs the rim is widest Inferiorly, then Superiorly, Nasally and Temporally. Glaucoma often breaks this order first at the inferior or superior pole.[10]

Lamina cribrosa

Eye & clinic

A sieve-like plate of collagen in the optic nerve head through which the axons leave the eye. Strain here from eye pressure is thought to injure them.

Ocular hypertension

Eye & clinic

IOP above 21 mmHg without damage to the optic nerve. It raises the risk of glaucoma, but many people with it never develop the disease.

Central corneal thickness

Eye & clinic

Goldmann tonometry is calibrated for an average cornea of about 520 µm. A thicker cornea reads falsely high and a thinner one falsely low.[4]

Tonometers

Eye & clinic

Goldmann applanation is the clinical reference; rebound and air-puff devices are also common. Readings differ between devices, one reason IOP doesn't transfer cleanly between datasets.

HRT (Heidelberg Retina Tomograph)

Eye & clinic

A confocal scanning laser device that measures optic disc topography: disc, rim and cup areas, cup volume and the linear cup-to-disc ratio. The LSMU morphometry uses these HRT-style parameters.[37]

Visual field and mean deviation (MD)

Eye & clinic

Perimetry measures light sensitivity at many points of the visual field. MD summarises the average deviation from age-matched normal values, in decibels; more negative is worse.

Antihypertensive drug classes

Eye & clinic

β-blockers, ACE inhibitors, angiotensin receptor blockers, calcium-channel blockers and diuretics lower blood pressure in different ways. GlaucomAI derives classes from drug names. Calcium-channel blockers have been trialled in NTG for their effect on blood flow.[38]

Topical β-blocker (timolol)

Eye & clinic

An IOP-lowering eye drop. Part of the dose reaches the bloodstream and can slow the heart and lower blood pressure, which matters for perfusion.

Mean arterial pressure (MAP)

Blood flow

The average arterial pressure over one heartbeat, estimated as diastolic + ⅓ × (systolic − diastolic).

Diastolic perfusion pressure (DOPP)

Blood flow

Diastolic blood pressure minus IOP: the pressure driving blood into the eye at the low point of each heartbeat. Values below 50 mmHg are linked with more glaucoma.[19]

Nocturnal hypotension

Blood flow

Blood pressure normally dips during sleep. A deep dip lowers perfusion at night, when IOP measured lying down can be higher; it has been linked with glaucoma progression.[22, 23]

Primary vascular dysregulation

Blood flow

Also called Flammer syndrome: a tendency of vessels to react inappropriately, with cold hands, low blood pressure and sometimes migraine. It is more common in NTG.[8]

Neurovascular coupling

Blood flow

Active neurons signal nearby vessels to widen so that blood flow matches demand. The flicker test uses this reflex to probe the retina's vessels.[26]

Pulsatility (Sonovum amplitude)

Blood flow

The LSMU cohort's pulsation measure, recorded with an ultrasound-based Sonovum device. What the amplitude physically represents is still being confirmed with the clinic, so GlaucomAI compares it only with age- and sex-matched controls.

Fundus photograph

Imaging & AI

A colour photograph of the back of the eye taken through the pupil. It shows the optic disc, the cup, the vessels and the macula.

B-scan and OCT volume

Imaging & AI

A B-scan is one OCT cross-section. A volume is a stack of B-scans covering an area such as the optic disc.[16]

Segmentation mask

Imaging & AI

An image-sized map that labels every pixel, here as disc, cup or background.

Dice coefficient

Imaging & AI

Overlap between two outlines: 2 × shared area ÷ (area A + area B). 1.0 is a perfect match and 0 means no overlap.[15]

Data augmentation

Imaging & AI

Random flips, rotations and brightness or colour changes during training, so the model learns the anatomy rather than one camera's look.

Test-time augmentation

Imaging & AI

Averaging a model's predictions over simple variants of the input. GlaucomAI averages the photo and its mirror image.

Plausibility gate (ungradable)

Imaging & AI

GlaucomAI refuses a photo when the disc it finds is implausibly small or large (outside 3–35% of the image's long side) or contains no cup, instead of guessing.

Expert disagreement band

Imaging & AI

Doctors outlining the same cup disagree too. The ± on the Try page is the typical spread between Chákṣu's five experts for a cup of that size.[12]

Phase correlation

Imaging & AI

Finds how far one image is shifted relative to another from the phase of their Fourier transforms. The flicker tracker uses it to cancel eye movements.[30]

Half-depth width

Imaging & AI

A vessel shows as a dark dip in the brightness profile across it. Its width is measured where the dip is half as deep, like a full width at half maximum.

Synthetic data and ground truth

Imaging & AI

Simulated video in which the true vessel diameter is known exactly, so the tracker's error can be measured before real recordings exist.

Percentile and tertile

Statistics & data

A percentile is the share of the cohort with a lower value. Tertiles cut the cohort into thirds at the 33rd and 67th percentiles; GlaucomAI's phenotype uses them.

z-score

Statistics & data

How many standard deviations a value lies from a reference mean: 0 is average, beyond ±2 is unusual.

Logistic regression

Statistics & data

Models the log-odds of disease as a weighted sum of inputs. GlaucomAI's screening model uses age, sex and IOP; coefficients per standard deviation show how strongly each input pulls.

Sensitivity and specificity

Statistics & data

Sensitivity is the share of patients a test flags; specificity is the share of healthy people it clears. Moving the threshold trades one for the other.[33]

Youden's J

Statistics & data

Sensitivity + specificity − 1. The threshold with the highest J is a common default operating point, and the one GlaucomAI's screening model uses.[34]

Bootstrap confidence interval

Statistics & data

Resample the patients with replacement thousands of times and recompute the metric each time; the middle 95% of the results is the interval.[35]

Holm correction

Statistics & data

When several hypotheses are tested, Holm's step-down method keeps the chance of any false positive at 5%. GlaucomAI has one primary question and corrects the rest.[39]

Grouped cross-validation

Statistics & data

Splitting data into folds by patient, not by eye, so the two eyes of one person never sit on opposite sides of a train/test split.

External validation

Statistics & data

Testing a finished model on data from another source: another hospital, camera or population. It is harder than an internal split and closer to real use.[36]

Pseudonymisation

Statistics & data

Replacing names and IDs with study codes, with the key kept separately. Under the GDPR, pseudonymised data is still personal data, which is why the patient pages can be switched off.

Provenance log

Statistics & data

A record of every value GlaucomAI excluded or corrected, with the rule and the reason, so each decision can be reviewed and reversed.

References on this page

  1. 4.Ehlers N, Bramsen T, Sperling S. Applanation tonometry and central corneal thickness. Acta Ophthalmol (Copenh). 1975;53(1):34–43. PubMed search ↗
  2. 8.Flammer J, Konieczka K, Flammer AJ. The primary vascular dysregulation syndrome: implications for eye diseases. EPMA J. 2013;4(1):14. PubMed search ↗
  3. 9.Jonas JB, Budde WM, Panda-Jonas S. Ophthalmoscopic evaluation of the optic nerve head. Surv Ophthalmol. 1999;43(4):293–320. PubMed search ↗
  4. 10.Harizman N, Oliveira C, Chiang A, et al. The ISNT rule and differentiation of normal from glaucomatous eyes. Arch Ophthalmol. 2006;124(11):1579–1583. PubMed search ↗
  5. 12.Kumar JRH, Seelamantula CS, Gagan JH, et al. Chákṣu: a glaucoma specific fundus image database. Sci Data. 2023;10:70. DOI ↗
  6. 15.Dice LR. Measures of the amount of ecologic association between species. Ecology. 1945;26(3):297–302. Google Scholar ↗
  7. 16.Huang D, Swanson EA, Lin CP, et al. Optical coherence tomography. Science. 1991;254(5035):1178–1181. PubMed search ↗
  8. 19.Bonomi L, Marchini G, Marraffa M, et al. Vascular risk factors for primary open angle glaucoma: the Egna-Neumarkt Study. Ophthalmology. 2000;107(7):1287–1293. PubMed ↗
  9. 22.Hayreh SS, Zimmerman MB, Podhajsky P, Alward WL. Nocturnal arterial hypotension and its role in optic nerve head and ocular ischemic disorders. Am J Ophthalmol. 1994;117(5):603–624. PubMed search ↗
  10. 23.Charlson ME, de Moraes CG, Link A, et al. Nocturnal systemic hypotension increases the risk of glaucoma progression. Ophthalmology. 2014;121(10):2004–2012. PubMed search ↗
  11. 26.Riva CE, Logean E, Falsini B. Visually evoked hemodynamical response and assessment of neurovascular coupling in the optic nerve and retina. Prog Retin Eye Res. 2005;24(2):183–215. PubMed search ↗
  12. 30.Kuglin CD, Hines DC. The phase correlation image alignment method. Proc IEEE Int Conf on Cybernetics and Society. 1975:163–165. Google Scholar ↗
  13. 33.Fawcett T. An introduction to ROC analysis. Pattern Recognit Lett. 2006;27(8):861–874. Google Scholar ↗
  14. 34.Youden WJ. Index for rating diagnostic tests. Cancer. 1950;3(1):32–35. PubMed search ↗
  15. 35.Efron B, Tibshirani RJ. An Introduction to the Bootstrap. Chapman & Hall/CRC; 1993. Google Scholar ↗
  16. 36.Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 2015;350:g7594. PubMed search ↗
  17. 37.Wollstein G, Garway-Heath DF, Hitchings RA. Identification of early glaucoma cases with the scanning laser ophthalmoscope. Ophthalmology. 1998;105(8):1557–1563. PubMed search ↗
  18. 38.Koseki N, Araie M, Tomidokoro A, et al. A placebo-controlled 3-year study of a calcium blocker on visual field and ocular circulation in glaucoma with low-normal pressure. Ophthalmology. 2008;115(11):2049–2057. PubMed search ↗
  19. 39.Holm S. A simple sequentially rejective multiple test procedure. Scand J Stat. 1979;6(2):65–70. Google Scholar ↗