The medical problem that Raphael Brüngel wants to solve with his doctorate is a very specific one. Nevertheless, his solution will represent significant progress for many other areas.
Raphael Brüngel's research focuses on wound care. Wounds can become very complex structures, especially when they have become chronic. They consist of up to three main tissue types (granulation, fibrin coating and necrosis), the variety of combinations of which probably exceeds that of the colors in Renaissance paintings. Each wound is unique. Their reliable classification is complex.
Their healing process is just as unique. Numerous individual and medically relevant factors play a role here - age, general condition, previous illnesses such as diabetes and many more. A wound that takes longer than three months to heal is considered chronic. Their treatment is particularly demanding, for example there is a risk of infection, which can be accompanied by tissue loss. Any circulatory disorders that may be present can further restrict the wound's ability to heal itself. Setbacks are part of everyday life, even with good care.
In practice, nurses do not have enough time to regularly examine every wound in detail and document the healing process. This is why AI-supported solutions are being developed to facilitate early stage detection, condition and tissue classification, progression analysis and documentation.
The sore point of the data records
However, solutions have so far been driving with the handbrake on. Their learned expertise is based on as many and varied photos of wounds with suitable information as possible. Data sets available for research are rare and of variable quality.
And unbalanced in terms of content: Certain manifestations, such as wounds with dead ("necrotic") tissue, are far too rarely represented for AI to reach its full potential. There is also a lack of darker skin types in medical data sets.
Neural networks
This is where Brüngel comes in. He wants to fill in these poorly populated areas of the data sets with the help of special neural networks: These highly developed technologies are called "Generative Adversarial Networks" (GANs) and are able to "think up" realistic-looking representations as well as "translate" real images into other representations - and for the use case of wound images, precisely those that are in short supply in the data sets. Such artificially generated images are referred to as "synthetic" in technical jargon. In order to create them, the corresponding GAN models must have learned and understood the nature of these underrepresented wounds. As magical as this sounds, it is ultimately down to earth: it all goes back to clever statistics.
This is what Brüngel does: on the one hand, he develops methods and strategies that enable GANs to create wound representations that meet the highest standards. On the other hand, he is researching the potential and limitations of this technology in the context of optimizing applications.
Other medical and non-medical fields can benefit from Brüngel's doctoral project, which will run until 2025. For example, the methodology could be directly transferred to general skin lesions, such as melanomas. Due to this thematic proximity, Brüngel's doctoral project is associated with the WisPerMed research training group, which focuses on personalized medicine for melanoma.
Explanatory video
This explanatory video (English with German and English subtitles) describes the research field of Raphael Brüngel, a doctoral student at the Graduate Center of Dortmund University of Applied Sciences and Arts. It was produced by him for the DART Symposium at Fachhochschule Dortmund(Opens in a new tab) .