University Hospital Freiburg

AI Faster than Ophthalmologists

29. April 2019, 13:52 Uhr   |  Iris Stroh

AI Faster than Ophthalmologists
© Universitätsklinikum Freiburg

The software developed at the University Hospital in Freiburg was just as good, but considerably faster, than the software developed by trained ophthalmologists when counting microscope images.

Counting cells in tissue is necessary for many diagnoses, but so far it has been very time-consuming. Now, eye specialists at the University Hospital in Freiburg have developed software that is able to analyse cell images of the cornea as precisely as a human - in a fraction of the time.

They provide a clear view: The endothelial cells in the eye continuously pump water out of the cornea and keep it transparent. If the cells die, the cornea becomes cloudy. In order to start treatment at the right time, ophthalmologists had to evaluate microscopic images of the endothelial layer and count the cells by hand. Scientists at the University Hospital of Freiburg have now developed self-learning software that is able to do this very reliably. "The self-learning software is able to do what even an experienced person needs several minutes to do," said Prof. Dr. Daniel Böhringer, head of clinical studies at the Department of Ophthalmology at the Freiburg University Medical Centre. The software study was published in the Open Access journal Scientific Reports, which is part of the Nature group.

Together with his team, Prof. Böhringer developed the evaluation method. The basis was the "U-Net" software developed at the University of Freiburg. The software is based on the classical approach of neuronal networks that have self-learning abilities. The study evaluated 385 microscope images of healthy and diseased eyes. Although the images varied greatly in quality, there was very good agreement between the automated image analysis and the hand-counted measurements. This was also the case for poor quality images. In addition, almost all non-evaluable images were marked as such.

"With the U-Net, we succeeded in automating the time-consuming task of tissue analysis. This new method greatly expands the research possibilities in ophthalmology," said Prof. Dr. Thomas Reinhard, Medical Director of the Department of Ophthalmology at the University Hospital of Freiburg. "This will enable us to re-evaluate archived images and investigate considerably more images in future studies. CE certification is required in order to use the software in the diagnosis of patients. The Freiburg researchers are currently investigating this.

At birth, approximately 3,000 to 5,000 endothelial cells can be counted per square millimetre. The cells gradually die with age. Only when more than 90 per cent of the endothelial cells have died as a result of disease or surgery does vision decline and severe pain occur. This can only be treated with a corneal transplant because the endothelial cells do not grow back. "If we recognise pathological changes in good time, we can often help the patients," said Prof. Reinhard.

Auf Facebook teilen Auf Twitter teilen Auf Linkedin teilen Via Mail teilen

Verwandte Artikel