In a new cross-sectional study published in JAMA Ophthalmology,1 researchers posed the question: can real-time surgical guidance for cataract surgery be achieved using a deep learning detection network combined with computer vision tools? The short answer: yes, it can. Computer vision is a newer approach to AI that allows an automated processing system to “see” the visual world and react accordingly.

In the study, researchers used a region-based convolutional neural network to track the pupil and identify the current surgical phase being performed with a mean area under the curve greater than 95%, triggering surgical guidance tools developed with computer vision.

The investigators say that the results suggest that an AI-based surgical guidance platform has the potential to enhance the surgeon experience in phaco.

Here are some of the key findings:

  • The system was able to perform precise pupil tracking and segmentation as well as surgical phase identification in real time during phaco.
  • Computer-vision tools were able to use the information retrieved by neural networks with the potential to provide surgical guidance to try to improve capsulorhexis symmetry, provide feedback for harmful turbulence and instrument movements, and improve tissue visualization.
  • A surgical guidance platform was able to use video output from existing commercially available surgical systems. This platform combined surgical-phase identification and element tracking within the same neural network in real time, also providing a framework for the development of additional guidance tools.

In an accompanying editorial,2 Amy Yuan, MD, and Cecilia S. Lee, MD, MS, say it’s “compelling” to train computers to learn surgery; it may bring clinical and educational benefits, especially since errors and complications that occur during surgery can result in patient morbidity. But they note that future challenges may await researchers and surgeons.

“Given the routine use of operating microscopes in ophthalmic surgery, vast quantities of surgical videos could be used to train precise deep learning algorithms, although devising a system to share surgical data, particularly labor-intensive, expert-annotated data, is an important challenge that will need to be addressed,” Drs. Yuan and Lee note.


1. JAMA Ophthalmol 2021; Jan. 13, 2022. [Epub ahead of print]

Nespolo Garcia R, Yi D, Cole E, et al.  

2. JAMA Ophthalmol 2021; Jan. 13, 2022. [Epub ahead of print]

Yuan A, Lee CS.


Reviewing Practice Guidelines in DME

Researchers identified diabetic macular edema clinical practice guidelines that made anti-VEGF treatment recommendations, and assessed their reporting quality and their considerations when making recommendations. Eligibility criteria included CPGs published between December 2009 and December 2019 that made explicit anti-VEGF recommendations in DME.

Sources of evidence included a sensitive search strategy in Embase, Google Scholar and hand-searching on 165 websites. Researchers extracted information from each CPG with a previously piloted sheet. Two independent authors applied the Appraisal of Guidelines, Research and Evaluation tool (AGREE-II) assessment for each CPG.

Here are some of the findings:

  • Twenty-one sources including CPGs recommended anti-VEGF for DME, with wide variation among the clinical aspects included, such as location of DME, visual acuity required, and therapeutic alternatives or discontinuation. 
  • Most had a poor quality of reporting based on the AGREE-II tool assessment, especially those developed by ophthalmological societies, those that had exclusive content about DME, and those in which most authors disclosed conflict of interests with pharmaceutical industry or where the authors didn’t report COIs. 
  • Pharmaceutical-sponsored CPGs didn’t use systematic reviews to support their recommendations. 
  • Very few recommendations considered patient values and preferences, equity, acceptability and feasibility of the intervention.

Researchers determined that most of the clinical practice guidelines that made recommendations of anti-VEGF for DME had poor quality of reporting, didn’t use systematic reviews and didn’t consider patients’ values and preferences. 


Br J Ophthalmol 2021; Dec. 14, 2021. [Epub ahead of print]

Vargas-Peirano M, Verdejo C, Vergara-Merino L, et al.


Outcomes in Neuropathic Ocular Pain Therapy

Recently, a team of researchers performed a study to more clearly determine the clinical responses to common treatment modalities in patients with neuropathic ocular surface pain.

The study included 101 patients with a clinical diagnosis of neuropathic pain who were seen at the University of Miami Oculofacial Pain Clinic between January 2015 and August 2021. Data from medical records including comorbid conditions and information on current or previous medication use or treatments was collected for each patient. The cohort was divided into the following subcategories based on the type of pain:

  • Post-surgical: pain that developed after undergoing surgery (e.g., refractive or cataract);
  • Post-traumatic: pain that followed a non-surgical trauma such as chemotherapy, radiation or brain injury;
  • Migraine-like: bilateral pain that started spontaneously and was accompanied by photophobia and, typically in this cohort, migraine or headache; and
  • Unilateral: pain that started spontaneously, did not follow surgery, was not typical for trigeminal neuralgia and had neuropathic qualities.

There were various treatment modalities used among the study population. The researchers wrote, “The most common oral medications were α2δ ligands (48.5 percent), nonsteroidal anti-inflammatory drugs (31.7 percent) and serotonin-norepinephrine re-uptake inhibitors (16.8 percent). Oral medications were commonly paired with topical therapy, such as autologous serum tears (20.8 percent) and/or a topical anti-inflammatory (e.g., topical steroid [19.8 percent], cyclosporine or lifitegrast [17.8 percent] or less commonly, tacrolimus [8.9 percent]). Finally, a minority of patients received adjuvant therapies, including trigeminal nerve stimulation (TNS, 15.8 percent), steroid/anesthetic-based periocular nerve block (24.8 percent) and/or botulinum toxin injections (10.9 percent).”

The researchers made the following conclusions from their analysis of the recorded clinical responses to each treatment method:

At least one oral medication reduced pain to a mild degree or greater for most patients in the post-traumatic (81.2 percent), migraine-like (73 percent) and unilateral pain (72.7 percent) groups but in only 38.5 percent of the post-surgical pain patients.

Marked improvement from treatment with oral medications was most frequently found in migraine-like patients (21.6 percent) and to a lesser degree in the other groups (post-surgical 15.4 percent, post-traumatic 12.5 percent, unilateral 0 percent).

The investigators note that topical medication subjectively improved pain in the post-traumatic (66.7 percent), migraine-like (78.6 percent) and unilateral (70 percent) groups more often than in the post-surgical group (43.7 percent). Marked improvement was also most common in migraine-like patients for this treatment (21.4 percent).

One or more adjuvants reduced pain to a mild degree or greater in the post-surgical (54.5 percent), post-traumatic (71.4 percent) and migraine-like (73 percent) groups but did so for none of patients in the unilateral group.

As is clear from the data, responses to a number of different treatment methods vary between pain subgroups. 

The researchers concluded that, though treatment responses vary by individual, different trends can be observed among pain subgroups. Future studies will hopefully identify ways to pinpoint the location(s) of nervous system dysfunction to decrease reliance on a trial-and-error therapeutic approach.


Front Pharmacol 2021. 12:788524.

Patel S, Mittal R, Felix ER, et al.


Long-term DMEK Outcomes

Scientists evaluated clinical outcomes of patients up to 10 years after they underwent Descemet’s membrane endothelial keratoplasty, as part of a retrospective, consecutive, single-center case series. 

The medical files of eyes that received DMEK between 2009 and 2012 for the treatment of endothelial dysfunction were evaluated for follow-up time and clinical outcomes. Scientists analyzed the annual exams of 66 eyes (with a minimum of eight years of follow-up) for best-corrected visual acuity, endothelial cell density (ECD) and central corneal thickness (CCT).

Here are some of the findings from the study: 

  • Best-corrected visual acuity improved from 0.55 ±0.37 logMAR (n=54) to 0.15 ±0.11 (n=47) in eyes without ocular comorbidities one year after DMEK (p<0.001) and remained stable up to 10 years after DMEK. 
  • Mean endothelial cell density in the patients decreased to 744 ±207 cells/mm2 (n=39) after nine years and to 729 ±167 cells/mm2 (n=21) after 10 years. 
  • The average central corneal thickness decreased from 650 ±67 µm before DMEK to 525 ±40 µm (n=56) after one year; and increased to 563 ±40 µm (n=39) after nine years and to 570 ±42 µm (n=21) after 10 years. 
  • After year eight, graft failure occurred in four eyes that required repeat DMEK after 101 to 127 months.

The researchers reported that visual acuity remained stable in spite of slowly increasing corneal thickness and diminishing endothelial cell density during the 10-year period after DMEK.


Int Ophthalmol; Jan. 8. [Epub ahead of print]

Weller JM, Kruse FE, Tourtas T.