EvoEye - Intelligent fundus layers detection based on Artificial Intelligence Agent
EvoEye is an Agent based on artificial intelligence, which is able to recognize and mark individual layers of the fundus on OCT eye scans. EvoEye agent is based on the model and the knowledge that gave him an expert. In fact, he needs a teacher who showed him how to mark the bottom of the eye.
Based on the knowledge provided by the teacher, EvoEye Agent learns the rules and dependencies that allow for the proper
recognition of the fundus layers.
EvoEye Agent is able to learn from many specialists. Thanks to this, he is able to have all their knowledge and remember
all medical cases he has learned at the same time. As a result, it is very easy for him to find solutions for new cases,
because he always remembers everything he has learned.
key features
- Recognizes 128 b-scans in about 25s on a weak computer
- Can be run on a computer in a doctor's office or as a remote service
- Perfectly seprarates choroid and vitreous
- Accurately marks small and medium blood vessels
It is possible to visualize layers as separate spaced apart segments to allow for more accurate analysis. This is shown in the image above.
It is very important that EveEye Agent AI is able to separate the vitreous and choroid from the rest of the image. This allows processing of the main area of interest which are the fundus layers. Both vitreous and choroid can also be processed separately.
The image above shows a case where a large blood vessel was detected as an area of uncertainty.
In our approach, this size of blood vessels should be treated as
a pseudo pathology in terms of the fundus continuity. Smaller blood vessels are marked correctly.
In contrast to artificial intelligence methods, deterministic algorithms would lead fundus layers lines in
an average or statistical way.
As a result, this introduces false data in the context of a correctly running layer of the fundus.
The advantage of an EvoEye Agent based on Artificial Intelligence is consistent decision making or uncertainty in specific areas.
This is much more valuable than averaging or statistical layering because it is more akin to human behavior.
The consequence is also noticeable in the uniform way of marking layers. Despite the twofold method of marking the layers by teacher
(once upside down, once downside down), Agent AI has developed a systematic way of marking them.
Red is the layer marked by EvoEye AI Agent, and blue is the layer marked by the teacher. This is shown in the image above.