Role Of Artificial Intelligence On How Macular Hole Is Measured By Swept-Source OCT

A form of OCT known as swept-source (SS) OCT offers increases in the ability to see the vitreous, retina, and choroid all at once.

SS-OCT is the best method for recording a variety of views and analyzing the structures beneath the retinal pigmented epithelium (RPE), particularly the choroid, because of its faster scan rates, reduced signal attenuation, more thorough imaging, and deeper tissue penetration.

Through a change in the integrity of the outer retinal hyperreflective bands, OCT can reveal the gradual degeneration of the photoreceptor layers.
Macular anomalies such CME, epiretinal membrane (ERM), aberrant vitreomacular traction interfaces, or macular holes can also be found with OCT.

However, only a few studies have been done to assess the relationship between VA and OCT alterations.

Previous research has demonstrated that in RP patients, the ellipsoid zone (EZ) was linked to improved VA and a thicker fovea. At the same time, people with RP may have foveal dysfunction if there is no EZ.

A presentation in the Lions' Lair competition, a competition for retina research proposals held during the 2021 Canadian Ophthalmological Society Annual Meeting and Exhibition, focused on the creation of a deep learning model to assist with segmentation of swept-source optical coherence tomography (OCT) images to measure the volume of macular holes.

The goal of this is to perhaps improve the 2013 classification of macular holes made by the International Vitreomacular Traction Study [IVTS] group.

By adopting an OCT-dependent and anatomical categorization of macular holes, the IVTS improved the first classification of macular holes. The categorization of macular holes using IVTS is essential for surgical planning and patient prognostic factors.

The IVTS group believed that the primary cause of macular holes was anatomical classification. Small macular holes are less than 250 m in diameter, medium macular holes are between 251 m and 400 m in diameter, and giant macular holes are greater than 400 m in diameter.

However, regions other than the macula can experience vitreomacular traction. The retina is a three-dimensional organ as well.

Surgery students and patients alike should not be forced to view the retina in two dimensions.

Although the retina is a 3-dimensional structure, utilizing a measurement such minimum linear diameter is in fact 1-dimensional.

Vitreomacular traction was easier to spot with 3-dimensional viewing. In order to find tiny full-thickness macular holes, high density radial scanning is more effective than traditional raster volume scanning.

Furthermore, studies have revealed significant inter- and intrauser variability for measures of macular holes with a minimum linear diameter within the same eye.

This alters prognostic factors and surgical planning, often moving a patient from being classified as having a small macular hole to having a big deficiency and vice versa.

There is a significant amount of OCT segmentation dependence with low linear diameter. The major biomarker for surgical planning for macular hole surgery is [minimum linear diameter], so our biomarkers must be as exact, repeatable, and accurate as feasible for our patients.

Automated measures of macular hole volume were obtained using a deep learning model, which the researcher defined as a reliable, three-dimensional biomarker for macular hole size that can help with surgical planning and prognosis counseling for patients with full-thickness retinal holes.
25 patients with full-thickness macular holes from the Vitreous Retina Macula Specialists of Toronto (VRMTO) practice participated in the pilot study. 3-dimensional OCT pictures of the holes were taken prior to surgery, one month after surgery, and one year following surgery.

The minimal linear diameter and macular hole volume were carefully determined by Choudhry, a single vitreoretinal surgeon who serves as the medical director at VRMTO. A convolutional neural network was created to automatically determine volume.

As compared to the minimum linear diameter, researchers also examined to determine if there were any links between volume and changes in eyesight.

Our deep learning model's automated estimation of the macular hole volume and our vitreoretinal surgery's manual estimation of the same parameter had a 0.94 correlation.

Before being used in routine surgical practice, the deep learning model needs to be validated. This would give vitreoretinal surgeons another tool in their toolbox for prognosticating patient outcomes and planning surgeries.

Increasing the number of OCT images in the training set will aid in determining the true size of the macular hole.

In the future, AI might offer knowledge such as how much of the internal limiting membrane should be peeled off and use information from previously performed procedures.