Contact Person

Daniel Blase, M.Sc.

Project Description

Three layers of skin form a barrier, which protects us and provides us with sensory information about our external environment. Open wounds depict breaches of that barrier, which normally follow an orderly wound healing process. If a wound stagnates during the process (1-3 months) - often at the inflammatory stage – despite proper treatment, it is referred to as a “chronic wound”. Especially the European health system faces two major problems with chronic wounds: Not only does the aging of a society fuel the frequency of occurrence of serious chronic wounds, e.g. of pressure-induced ulcers in geriatric patients, but their prevention and treatment are relatively costly and difficult, too. The current diagnosis is largely based on educated guesses after a haptic and visual inspection, and on traditional measurement tools, such as disposable rulers, for wound surveying. This process often fails to detect chronic wounds in early stages and makes it hard to track the size of the wound margin for example, which is an important measure for its state. For a more reliable prediction, a more robust, objective and automated feature extraction, analysis and interpretation are required. In addition, contact-less measurement techniques are considered beneficial to the wound healing process, as they would prevent further irritations. Last but not least, effortless tracking of conclusive wound parameters over time could support timely treatment and reduce stress for the patient during the healing process.
Combining temperature distribution measurements with traditional, visual features, this project will include the design of a small, mobile, multi-modal type of “scanner”, which will feature one infrared thermology camera, as well as multiple RGB and/or MONO cameras for color evaluation and depth vision. The diagnostic process will include image collection, inference of wound parameters as well as the short-time analysis based on dedicated deep learning (DL) algorithms. Furthermore, a safe and smooth image transmission onto a typical hospital’s servers and into a conclusive, digital patient file for long-term analysis of the wound development will be included in this project.
Especially the short-term inference and analysis tasks will be performed by the brain of the “scanner”, whose software will be tailored to run on a to-be-designed “neuromorphic hardware” platform, which will be solely designed for a fast and energy-efficient implementation of neural networks, allowing mobile yet complex applications.

Project Goals

• Mobile, miniaturized camera setup for contactless wound assessment
• Inference of conclusive wound parameters using deep learning (short-term analysis)
• Comparison with previous wound measurements using deep learning (long-term analysis)
• Tailoring the software to the “neuromorphic hardware”

Project Partners

STAR Healthcare Management GmbH
Gremse IT GmbH