ARTIFICIAL INTELLIGENCE IN DIAGNOSIS OF GLAUCOMA

Over the last few years there has been a spurt in new technologies to improve diagnosis of glaucoma. However, these methods (such as Optical Coherence Tomography, automated perimetry etc)  still lack the sensitivity and specificity required to diagnose glaucoma in all practice scenarios. In this situation AI is stepping in to make glaucoma diagnosis faster, easier and more reliable. 

A number of studies, including those being performed by Zhongshan Ophthalmic Center, attempt to combine clinical data of glaucoma patients with AI techniques to create programs that can screen and diagnose glaucoma. The trial, registered with the Clinical Trials Network is called Artificial Intelligence assisted Glaucoma Evaluation (AGE).

https://clinicaltrials.gov/ct2/show/NCT03268031 

The OCTAGON (Optical Coherence Tomography and Artificial Intelligence for Glaucomatous Optic Neuropathy) Research Group in the UK aims to impact glaucoma diagnosis and prognosis using AI (deep learning) and 3D medical imaging (OCT). To simplify glaucoma management the team is developing custom-written AI algorithms that can be applied to 3D OCT images of the optic nerve head (ONH).

http://www.bioeng.nus.edu.sg/oeil/OCTAGON.html 

Techniques have also been developed to “denoise” OCT images of the ONH using a Deep Learning approach. In a study reported in September 2018, an algorithm was successfully used to denoise single-frame OCT B-scans. These denoised B-scans were qualitatively similar to their corresponding multiframe B-scans with enhanced visibility of the ONH tissues. 

https://arxiv.org/abs/1809.10589 

The same team (reported above) also developed DRUNET: A dilated residual U-net Deep Learning network to digitally stain (i.e highlight) 6 ONH tissue layers by capturing both the local (tissue texture) and contextual information (spatial arrangement of tissues). As ONH tissues exhibit complex changes in their morphology with the progression of glaucoma, their simultaneous isolation may be of great interest for the clinical diagnosis and management of glaucoma.

https://www.osapublishing.org/boe/abstract.cfm?uri=boe-9-7-3244 

Visulytix announced its Pegasus-disc, an AI decision support tool for the analysis of retinal images to be equivalent to human experts in glaucoma detection. The findings of a study based on the technology was presented at the ARVO meeting held in Honolulu, Hawaii in April 2018.

https://visulytix.com/ 

Artificial Neural Networks have also been used to evaluate the relationship between macular vessel density and ganglion cell to inner plexiform layer thickness (GCIPLT) in glaucoma patients. The study found that MVD was significantly decreased in glaucoma patients and showed an almost linear correlation with macular GCIPLT. The ANN improved the performance of assessing macular GCIPLT.

https://www.practiceupdate.com/content/new-approaches-in-glaucoma-diagnosis-role-of-artificial-intelligence/70911

A group of Spanish researchers has developed an algorithm named RetinaLyze Glaucoma, which compares the color of the optic nerve to the color of the veins and arteries of the retina and deduces its degree of perfusion. The software uses AI to 24 sectors of the ONH to obtain a mathematical function which defines the normal eye by looking at the distribution of the Blood Globin Protein (GDF) and detects abnormalities such as those seen in glaucoma.

https://www.retinalyze.com/post/retinalyze-new-ai-algorithm-detects-glaucoma-using-retinal-images 

In November 2018, George and Matilda Eyecare and IBM Research Australia announced a joint collaboration to study fundus photos, OCT images and the use of AI algorithms to inform and potentially help guide practitioners in detecting glaucoma.

https://www.glaucoma.org.au/articles/george-matilda-ibm-collaborate-on-ai-for-glaucoma-article/ 

Extreme Learning Machine (ELM) and Fractal feature analysis have also been reported to detected glaucoma. In a letter to the International Journal of Ophthalmology, Kavitha et al presented their study and discussed various other studies which utilized ELM and fractal geometry to discriminate glaucomatous eyes from normal images.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4651900/ 

Silva et al have used machine learning classifiers such as Bagging (BAG), Naive-Bayes (NB), Multilayer Perceptron (MP), Radial Basis Function (RBF), Random Forest (RF), Ensemble Selection (ENS), Classification Tree (CTREE), Ada M1 (ADA), Support Vector Machine Linear (SVML) and Support Vector Machine Gaussian (SVMG) in glaucoma diagnosis. The study concluded that MLCs trained on OCT and SAP data can successfully discriminate between healthy and glaucomatous eyes. Also, the combination of OCT and SAP measurements improved the diagnostic accuracy compared with OCT data alone.

https://www.ncbi.nlm.nih.gov/pubmed/23929078