Artificial Intelligence and Deep Learning as the Future of Glaucoma Diagnosis

Limitations in Early Diagnosis

Early and correct diagnosis can lead to slowing the progression of glaucoma and prevent future vision loss. Detection of glaucoma is most challenging in the early stages of the disease. The primary issue in detecting it is that the most noticeable symptoms appear only after significant vision loss has occurred.

Therefore, current examinations should soon be enhanced with modern automated systems working on the principles of machine learning and artificial intelligence (AI), which will shorten the diagnostic process and minimize possible inaccuracies caused by a physician's subjective assessment through precise and rapid image interpretation. Currently, glaucoma diagnosis is based on several sub-examinations and relies on the doctor's skill and experience. Previous research has shown that even experienced ophthalmologists underestimate the probability of glaucoma in up to 20% of evaluated optic disc images.

We Already Have the Data for Machine Learning

The first procedures utilizing principles of artificial intelligence for comparing retinal images were proposed in the mid-20th century. In the 1990s, deep learning models based on neural networks principles were also first used for glaucoma diagnosis.

Thanks to a long history and technological progress, extensive datasets of fundus photographs are now available, which, due to the development of computer systems and algorithms, can be used for automated and objective interpretation of individual eye images.

Comparison of Different Methods

The authors of the study decided to compare different systems and algorithms. They categorized the models into three groups based on their operating principles. They distinguished models of artificial intelligence, machine learning, and deep learning. Artificial intelligence works by mimicking human behavior, machine learning through analyzing a certain dataset and past experiences. Deep learning, as the most advanced process in computer evolution, operates based on calculations using multi-layered neural networks and autonomous decision-making.

Based on the comparison and testing of individual models, the study authors consider methods based on deep learning principles to be the most suitable. They show much more accurate results in tasks such as image processing, pattern recognition, and subsequent diagnosis and prognosis of glaucoma.

In recent years, several innovative deep learning models specifically for glaucoma diagnosis have been developed, showing promising results. However, none have yet been approved by the U.S. Food and Drug Administration (FDA) for clinical practice use. Scientists believe this may be due to inconsistencies in the definition of glaucoma or the generalizability and interpretability level of individual models. To better integrate new technologies into healthcare facilities, further research should focus on these issues and create new standards for legal and ethical questions.

(jko)

Sources: 
1. Huang X., Islam M. R., Akter S. et al. Artificial intelligence in glaucoma: opportunities, challenges, and future directions. BioMed Eng Online 2023 16 Dec; 22: 126, doi: 10.1186/s12938-023-01187-8.
2. Yousefi S. Clinical applications of artificial intelligence in glaucoma. J Ophthalmic Vis Res 2023; 18 (1): 97–112, doi: 10.18502/jovr.v18i1.12730.