Development of "Predict ME," an online classifier to aid in differentiating diabetic macular edema from pseudophakic macular edema.

Hecht, Idan; Achiron, Ran; Bar, Asaf; Munk, Marion; Huf, Wolfgang; Burgansky-Eliash, Zvia; Achiron, Asaf (2020). Development of "Predict ME," an online classifier to aid in differentiating diabetic macular edema from pseudophakic macular edema. European journal of ophthalmology, 30(6), pp. 1495-1498. Sage 10.1177/1120672119865355

Text 1120672119865355.pdf - Published Version Restricted to registered users only Available under License Publisher holds Copyright. Download (165kB) | Request a copy

PURPOSE

Differentiating the underlying pathology of macular edema in patients with diabetic retinopathy following cataract surgery can be challenging. In 2015, Munk and colleagues trained and tested a machine learning classifier which uses optical coherence tomography variables in order to distinguish the underlying pathology of macular edema between diabetic macular edema and pseudophakic cystoid macular edema. It was able to accurately diagnose the underlying pathology in 90%-96% of cases. However, actually using the trained classifier required dedicated software and advanced technical skills which hindered its accessibility to most clinicians. Our aim was to package the classifier in an easy to use web-tool and validate the web-tool using a new cohort of patients.

METHODS

We packaged the classifier in a web-tool intended for use on a personal computer or mobile phone. We first ensured that the results from the web-tool coincide exactly with the results from the original algorithm and then proceeded to test it using data of 14 patients.

RESULTS

The etiology was accurately predicted in 12 out of 14 cases (86%). The cases with diabetic macular edema were accurately diagnosed in 7 out of 7 cases. Of the pseudophakic cystoid macular edema cases, 5 out of 6 were correctly interpreted and 1 case with a mixed etiology was interpreted as pseudophakic cystoid macular edema. Variable input was reported to be easy and took on average 7 ± 3 min.

CONCLUSION

The web-tool implementation of the classifier seems to be a valuable tool to support research into this field.

Interest & Impact

Downloads

2 since deposited on 13 Aug 2019
0 in the past 12 months

Citations

5 Citations in Web of Science ®
6 Citations in Scopus

Search

in Google Scholar™

Services

Actions (login required)

Edit item

Actions (login required)

Edit item