Semantic segmentation of gonio-photographs via adaptive ROI localisation and uncertainty estimation

Abstract

ObjectiveTo develop and test a deep learning (DL) model for semantic segmentation of anatomical layers of the anterior chamber angle (ACA) in digital gonio-photographs.Methods and analysisWe used a pilot dataset of 274 ACA sector images, annotated by expert ophthalmologists to delineate five anatomical layers: iris root, ciliary body band, scleral spur, trabecular meshwork and cornea. Narrow depth-of-field and peripheral vignetting prevented clinicians from annotating part of each image with sufficient confidence, introducing a degree of subjectivity and features correlation in the ground truth. To overcome these limitations, we present a DL model, designed and trained to perform two tasks simultaneously: (1) maximise the segmentation accuracy within the annotated region of each frame and (2) identify a region of interest (ROI) based on local image informativeness. Moreover, our calibrated model provides results interpretability returning pixel-wise classification uncertainty through Monte Carlo dropout.ResultsThe model was trained and validated in a 5-fold cross-validation experiment on ~90% of available data, achieving ~91% average segmentation accuracy within the annotated part of each ground truth image of the hold-out test set. An appropriate ROI was successfully identified in all test frames. The uncertainty estimation module located correctly inaccuracies and errors of segmentation outputs.ConclusionThe proposed model improves the only previously published work on gonio-photographs segmentation and may be a valid support for the automatic processing of these images to evaluate local tissue morphology. Uncertainty estimation is expected to facilitate acceptance of this system in clinical settings.

Applications of Artificial Intelligence and Deep Learning in Glaucoma

“…However, gonioscopy often has a steep learning curve and agreement between different graders is quite poor, which immensely restricts the use of gonioscopy in clinical practice 114,115. Several researchers have attempted to apply AI to goniophotographs to assist clinical ophthalmologists in the diagnosis of glaucoma 116–128…”

“…Peroni et al123 performed semantic segmentation of the anatomic structure of the ACA with a DL system,129 achieving ~88% of average pixel classification accuracy in a 5-fold cross-validation on a very limited size annotated image data set. Subsequently, in 2021, Peroni et al128 continued to develop and test a new DL model. The new model achieved an average segmentation accuracy of ~91% in the test set 128…”

“…Subsequently, in 2021, Peroni et al128 continued to develop and test a new DL model. The new model achieved an average segmentation accuracy of ~91% in the test set 128…”

“…All the studies utilizing goniophotograph-based AI had a relatively small sample size and overly homogeneous distribution of subjects 116–121,123,124,128. This may be due to the difficulty of collecting high-quality gonioscopic photographs, a key challenge for this modality 115,130,131.…”

“…114,115 Several researchers have attempted to apply AI to goniophotographs to assist clinical ophthalmologists in the diagnosis of glaucoma. [116][117][118][119][120][121][122][123][124][125][126][127][128] In 2010, Cheng et al 122 proposed an automated system that could automatically differentiate between open angle glaucoma and angle closure glaucoma on goniophotographs, which achieved a specificity and sensitivity of 92.6% and 97.8%, respectively, based on the combination of all goniophotographic images from each patient's eye. Cheng et al 125 then developed a system for automatic ACA grading through the focal edge.…”