Free Papers

Quantifying graft detachment after Descemet’s membrane endothelial keratoplasty with deep convolutional neural networks

Free Paper Details

First Author: F.Heslinga THE NETHERLANDS

Co Author(s):    M. Alberti   J. Pluim   J. Cabrerizo   M. Veta           

Abstract Details

Purpose:

Knowing the degree of graft detachment after Descemet’s Membrane Endothelial Keratoplasty (DMEK) is crucial for post-operative decision making. However, precise quantification of graft detachment cannot be performed with the currently available software for Anterior Segment Optical Coherence Tomography (AS-OCT) scans. We developed a method to automatically locate and quantify graft detachment after DMEK in AS-OCT.

Setting:

1280 AS-OCT B-scans were collected as part of a randomized controlled trial conducted at the Department of Ophthalmology, Rigshospitalet - Glostrup. The randomized study included patients with Fuchs' endothelial dystrophy or pseudophakic bullous keratopathy eligible for DMEK surgery. Post-DMEK graft detachments were manually annotated by a DMEK-expert in all scans.

Methods:

Using the annotations, a deep learning pipeline was developed to localize scleral spur, center the AS-OCT B-scans and segment the detached graft sections. Detachment segmentation model performance was evaluated per B-scan by comparing (1) length of detachment and (2) horizontal projection of the detached sections with the expert annotations. Horizontal projections were used to construct graft detachment maps. All final evaluations were done on a test set not used during training of the models. A second DMEK expert annotated the test set to determine inter-rater performance.

Results:

Mean scleral spur localization error was 0.155 mm, whereas the inter-rater difference was 0.090 mm. The estimated graft detachment lengths were within a 10-pixel (~150μm) distance from the ground truth in 69% of the cases (77% for the second DMEK expert). Dice scores for the horizontal projections of all B-scans with detachments were 0.896 and 0.880 for our model and the second DMEK expert respectively.

Conclusions:

To the best of our knowledge, our deep learning pipeline is the first method to automatically and instantly localize graft detachment in AS-OCT B-scans. Horizontal detachment projections can be determined with the same accuracy as a human DMEK expert, allowing the construction of accurate graft detachment maps. Automated localization and quantification of graft detachment can become key to standardize clinical decision making and can support future research in the field.

Financial Disclosure:

-