Professional association of ophthalmologists in Germany. registered association
Dusseldorf (ots)
It is a term that is appearing more and more frequently and that sometimes causes conflicting feelings: artificial intelligence (AI). On the one hand, it is praised as a miracle tool for getting a grip on the challenges of the future, on the other hand, it also scares some people: Something is happening with sometimes very personal data that is beyond the control of those affected. There are already AI applications in medicine and especially in ophthalmology, which can be seen as important diagnostic support. They do not replace the (ophthalmological) medical assessment and decision, but they offer important assistance.
What is artificial intelligence anyway?
“Artificial Intelligence” is a subfield of computer science that involves designing computers and computer programs in such a way that they work on problems independently – similar to how humans would do this. One sub-area of AI is “machine learning”: computers can develop and constantly improve algorithms in order to evaluate large amounts of data. Another area of AI, “deep learning”, goes even further: It imitates biological structures of the nervous system (neural networks). Rapid advances have been made here in recent years, and in image recognition in particular there are applications that come close to human abilities to analyze images, and in some cases even surpass them. Thanks to AI, statements about a person’s age, smoking habits and blood pressure can be made simply by analyzing photos of the fundus of the eye.
Why is ophthalmology ideal for digitization?
AI applications are helpful where clearly defined questions encounter large amounts of data that contain the necessary information.
Ophthalmology is a discipline that has long used the opportunity to look inside the eye: Hermann von Helmholtz invented the ophthalmoscope in 1850, which is still an indispensable diagnostic tool for ophthalmologists today. Since then, more and more accurate and finer methods have been developed to map the structures of the eye.
Ophthalmologists are no longer satisfied with just looking into the eye. Where in earlier decades conspicuous findings were still recorded in hand-drawn sketches, digital retinal photos are now being taken. There are also other methods: Optical coherence tomography (OCT) and other laser scanning methods, for example, coupled with modern computer technology, offer ever new possibilities for examining and imaging the finest structures of the retina and optic nerve, for example. In a non-contact examination, which only takes a few seconds or minutes, images are created that even show individual cell layers. You should be aware that this is not analog imaging. The images that are created here are digital data made visible by the computer.
In this way, ophthalmologists produce large amounts of data during their examinations using modern methods, which provide precisely the material that can be evaluated with AI applications.
Why could AI play a role in glaucoma diagnostics?
Glaucoma is a complex eye disease in which many different factors play a role. Gradually, fibers of the optic nerve are lost. The result is visual field defects that progress if the glaucoma is not treated. An important risk factor is intraocular pressure. If it is too high, the nerve at the optic nerve head, i.e. at the point where it leaves the eye, comes under pressure. The blood supply to the nerve suffers and the nerve fibers die off.
A whole range of procedures have been established in glaucoma diagnostics for decades, including measuring the intraocular pressure and analyzing the visual field. In addition, there is OCT and its further development, OCT angiography (OCT-A), with which several areas around the optic nerve that are important for glaucoma can be examined:
- The peripapillary vascular density – i.e. the density of the blood vessels around the optic nerve head – can be displayed. This makes it possible to assess how well the optic nerve is supplied with blood.
- Also of interest is the condition of small vessels in the choroid of the eye adjacent to the optic disc (parapapillary choroidal microvessels).
- Around the optic nerve head, the thickness of the retinal nerve fiber layer (RNFL) can be determined using OCT.
- Other important parameters are the vessel density around the macula – the point of sharpest vision in the eye – and the complex of ganglion cells (Ganglion Cell Complex, GCC), which consists of the retinal nerve fiber layer, the ganglion cell layer and the inner plexiform layer.
Closely meshed intraocular pressure measurements, visual field analyzes and OCT findings enable detailed diagnostics, with which the course of this chronic disease and the success of the treatment must be precisely documented over the years. This creates enormous amounts of data. Evaluating them is becoming increasingly difficult and time-consuming. There is a risk of not seeing the forest for the trees.
Automated evaluation of the data can provide crucial help here – and working groups around the world are looking for ways to make AI usable for glaucoma diagnostics. A search on the English-language meta-database “PubMed” shows that the number of publications dealing with AI and glaucoma has skyrocketed in recent years.
How do computers learn?
What steps are necessary for computers to learn to analyze findings? When we see a barcode or a QR code, we only see disordered pixels that we cannot interpret. But we know that computers – for example our smartphones – can read the information hidden in the pixels and make it usable for us, for example by opening a link to a website.
In order to use AI in ophthalmology, the computers must now learn how to interpret OCT images, for example. To do this, people first analyze the raw data and convert it into maps that contain simple patterns. The OCT image is thus “translated” into a barcode that the computer can read. This happens hundreds of times. This is how a training dataset is created. This first step of “supervised machine learning” (checked or accompanied machine learning) is followed by the next: a computer algorithm (a set of instructions for solving problems) uses the training data set to evaluate image data itself on this basis. The computer can do the analysis much faster than a human could. Research groups that make use of such an AI solution can then include many more cases in their analyses. Research papers are already being published today that take into account up to 15,000 visual field analyzes or 20,000 OCT findings.
Which applications can patients benefit from?
Some examples of publications from the past few years show the first successful applications: AI can already evaluate visual fields – it even recognizes defects more reliably than human experts. Appropriate applications can thus support glaucoma diagnosis (1). It is even possible to predict how the visual field will develop: a working group had a recurrent neuronal network evaluate five visual field findings from patients and then predict how a sixth examination would turn out. The result was superior to conventional methods (2). Another application is the assessment of retinal nerve fiber layer thickness from fundus photographs. This application was trained on the basis of fundus images and RNFL measurements with OCT. The program can use the photos to distinguish well whether a rapid or moderate loss of nerve fibers is to be expected. It thus helps ophthalmologists in the long-term follow-up of the glaucoma disease and offers support for therapy decisions even where an OCT examination is not possible (3).
Conclusion
Artificial intelligence can be used well where clear questions meet large amounts of data that contain the necessary information. Ophthalmology has various procedures with which the finest structures inside the eye can be examined and imaged. This creates large amounts of data that can be evaluated with the help of AI. In ophthalmology and especially in glaucoma diagnostics, there are already some examples of how AI can be used in research and in patient care. AI is a tool, but by no means a substitute for ophthalmological expertise.
Prof. Dr. Hagen Thieme
University Hospital Magdeburg
Clinic and Polyclinic for Ophthalmology Leipziger Str. 44 (House 60b)
39120 Magdeburg
Phone: 0391-67-13571
Fax: 0391-67-13570
Email: [email protected]
Sources:
- Li F, Wang Z, Qu G et al. Automatic differentiation of glaucoma visual field from non-glaucoma visual field using deep convolutional neural network. BMC Med Imaging 18, 35 (2018).
- Park K, Kim J & Lee J Visual Field Prediction using Recurrent Neural Network. Sci Rep 9, 8385 (2019).
- Medeiros F, Jammal A & Mariottoni E Detection of Progressive Glaucomatous Optic Nerve Damage on fundus Photographs with Deep Learning, Ophthalmology 2021 Mar; 128(3): 383-392, https://doi.org/10.1016/j.ophtha.2020.07.045
Press contact:
Professional association of ophthalmologists in Germany
Tersteegenstr. 12
D-40474 Dusseldorf
Tel. 0211 – 4 30 37 00
E-mail contact for press inquiries: [email protected]
www.augeninfo.de
www.aad-kongress.de/pressekonferenzen/pressekonferenz-2023/
Original content from: Professional association of ophthalmologists in Germany. eV, transmitted by news aktuell
