• COVID-19
  • Biosimilars
  • Cataract Therapeutics
  • DME
  • Gene Therapy
  • Workplace
  • Ptosis
  • Optic Relief
  • Imaging
  • Geographic Atrophy
  • AMD
  • Presbyopia
  • Ocular Surface Disease
  • Practice Management
  • Pediatrics
  • Surgery
  • Therapeutics
  • Optometry
  • Retina
  • Cataract
  • Pharmacy
  • IOL
  • Dry Eye
  • Understanding Antibiotic Resistance
  • Refractive
  • Cornea
  • Glaucoma
  • OCT
  • Ocular Allergy
  • Clinical Diagnosis
  • Technology

Cutting-edge neuro-ophthalmology: Combining artificial intelligence, eye tracking

Publication
Article
Digital EditionOphthalmology Times: October 15, 2020
Volume 45
Issue 17

Technology with potential to improve treatment and evaluation of cortical visual impairment.


This article was reviewed by Melinda Chang, MD

Artificial intelligence (AI) is slowly being integrated into ophthalmology and other medical fields.

“In machine-guided medical research, a specific research question is not needed as in traditional medical research, because the research can be data-driven,” said Melinda Chang, MD, an attending physician in the vision center at Children’s Hospital Los Angeles and an assistant professor of clinical ophthalmology at the Keck School of Medicine at the University of Southern California. “Analysis with machine learning can predict new data, identify new patterns, or learn the best action.”

AI, Chang explained, is useful in medicine for classification and prediction, for example, for classifying images as they correspond to diseased persons or controls.

Related: Artificial intelligence in medicine: The good, the bad, and the ugly

AI in neuro-ophthalmology
The technology can detect papilledema from fundus photographs and characterize visual function and ocular motility in nonverbal individuals via eye tracking, Chang said.

She cited a recent study1 in which 15,846 fundus photos (14,341 for training and validation and 1505 for external testing) were obtained with digital cameras at 24 centers nationally and internationally.

An AI model was developed to classify the images as normal, papilledema, or a nonpapilledema abnormality of the optic disc.

The study found that the accuracy of the model for differentiating papilledema from nonpapilledema images was 87.5%, the sensitivity was 96.4%, and the specificity was 84.7%. When the accuracy was compared between normal images and those of any optic disc abnormality, the accuracy was 91.8%.

The study’s authors concluded that this model was most useful for identifying any optic disc abnormality and could be used in emergency departments or neurosurgery clinics to rule out papilledema without consulting with ophthalmology, Chang recounted.

Related: Teaching AI algorithms to identify corneal pathology: The future is now

Eye tracking
In an eye-tracking system, while a patient is viewing a visual stimulus on a computer monitor, an infrared camera records the corneal and pupillary light reflexes from each eye, which enables the direction of gaze to be calculated at any time point with high spatial and temporal resolution, she said.

Eye tracking has been studied to assess vision in nonverbal patients, eg, pediatric patients, those who are impaired neurologically, and those with developmental disorders.Its capabilities in visual rehabilitation and for measuring the degree of strabismus also have been assessed.

Chang focused on children with cortical visual impairment (CVI), ie, vision loss due to damage to the post-geniculate visual pathways in the brain.

“This is the most common cause of pediatric visual impairment in the US and developed countries,” she said.

Related: Blockchain technology aims to drive big data to 'train' AI

CVI risk factors include prematurity, hydrocephalus, seizures, hypoxic ischemic encephalopathy, and other neurologic conditions.

Children with CVI have visual deficits that may differ from children with ocular causes of visual impairment.

For example, CVI causes more difficulty with low contrast; motion sensitivity may be spared in some children with CVI; and when in environments that are visually complex, children with CVI may have more difficulty seeing, she explained.

“Having an understanding of these unique visual deficits is important when guiding therapy. We know that these patients have the capacity to improve over time, but the factors that drive improvement are unclear,” Chang said.

Visual stimulation therapy is widely used, but no controlled trials have been conducted to assess the efficacy of visual stimulation.

“Establishment of an objective method for assessing visual stimulation (and other therapies) is important for pediatric patients with CVI,” she emphasized.

Related: AI screening: Allowing physicians to see eye more clearly from afar

Evaluating CVI with eye tracking and AI
Chang demonstrated patterns of fixation in children with CVI using these technologies. A child who served as a control was seen to be fixating on a baby’s eyes and hands in a brightly colored photo of that baby in a ball pit.

In contrast, a child with CVI was seen to be fixating mostly on the brightly colored balls with fewer fixations on the baby’s face in the same photo.

“The child with CVI appears to prefer looking at bright colors rather than faces,” she commented.

Problems with low contrast were demonstrated in a second substantially darker photo in which a child with CVI fixated only once on a low-contrast object (bird) compared with a normal child who focused numerous times on a low-contrast human face in addition to the bird.

Another type of visual deficit was demonstrated in a photo with a highly complex background in which a child with normal vision had many fixation points in contrast to a child with CVI who had no fixation points within the complex image; the child with CVI had many more fixation points when the picture was significantly simplified with a plain background.

Related: Making the most of imaging in examination of pediatric patients

“These findings demonstrate the crowding effect in CVI, in which vision is worsened by visually complex scenes and distractors,” Chang explained.

Without machine learning, Chang and associates were able to do simple analysis to determine that children with CVI had several visual deficits compared to controls, such as decreased contrast sensitivity and increased saccadic latency.

The investigators will use AI combined with eye tracking to identify other visual and oculomotor abnormalities, and potentially identify new deficits specific to CVI.

“The model will help determine the relative importance of various features such as color, background, faces, motion, and contrast, among others, in facilitating vision in children with CVI,” she said.

Related: The challenges of examining visual fields in children

Any information gained from such an analysis will help identify therapeutic targets and features that can be used to monitor disease severity and treatment efficacy.

“AI is a powerful tool with the potential to automate image analysis for neuro-ophthalmic conditions. Eye tracking facilitates visual assessment in nonverbal individuals, including children with CVI,” Chang concluded.

Read more by Lynda Charters


--

Melinda Chang, MD
e:mchang@chla.usc.edu
Chang has received grant funding from the Children’s Eye Foundation of AAPOS and the Knights Templar Eye Foundation.

--


Reference

1. Milea D, Najjar RP, Jiang Z, et al. Artificial intelligence to detect papilledema from ocular fundus photographs. N Engl J Med. 2020;382:1687-1695. doi: 10.1056/NEJMoa1917130

Related Videos
© 2024 MJH Life Sciences

All rights reserved.