Medical Image Processing

Intelligent Processing of Medical Image Data

The majority of our projects focus on the processing of medical image data (e.g., MRI or CT scans). These tools can be easily integrated into your infrastructure and provide the opportunity to facilitate the diagnostic process of specific diseases by performing automatic segmentation and highlighting abnormalities in medical images. It is important for us to leave the final decision in the diagnostic process to the treating medical experts in order to ensure the highest accuracy and safety for the patient. In addition to allowing considerable time savings to be realized, the synergy between humans and machines reduces the workload placed on medical professionals.

Detecting Rare Diseases and Connecting Patients

In the project »unrare.me«, we are collaborating with the University Hospital Bonn (UKB) to develop a platform to connect between patients with rare diseases. This addresses the significant challenges faced by individuals with rare diseases in finding specialized doctors with experience in treating their specific condition. The search is extremely difficult and often involves many hours of driving - or, in the worst case, comes to nothing.

The core idea here is to bring patients (with their consent) with similar medical conditions into contact so that they can share their medical history and treatment options as well as contacts to specialists. Another positive outcome is that, with the successful scaling of this project, the data of patients with the same rare disease can be aggregated in a database which can significantly simplify the identification of participants for disease casuality.
 

• More info (in German)

AI-Supported Diagnostics in the Radiology Department of the University Hospital Bonn (UKB)

In collaboration with the University Hospital Bonn (UKB), we have developed an AI-based assistance system as part of a project led by Professor Dr. Attenberger. This system provides support in diagnosing the four most common thoracic diseases and determining the position of foreign bodies.

With the aim of being able to identify these four diseases, we first had to solve the problem of the scarce availability of large and high-quality data sets required for training our AI/ML models at the start of the project. The classification is typically based on ICU chest X-rays, which are done by medical experts in text form as findings.

While these medical findings theoretically offer an extensive database and a solid starting point for developing AI and ML solutions, the findings are usually not standardized or presented in a sufficiently structured form for the training of our tools. To address this issue, we initially trained our Natural Language Processing algorithm to classify the unstructured reports. Subsequently, we trained our model to learn the relationship between a report and the corresponding image to ultimately enable automatic image evaluation.

Training the Algorithm for Diagnostic Support

Immunological disorders are currently predominantly treated symptomatically with medications that non-specifically suppress the immune system of patients (immunosuppressive therapy). A causal, personalized therapy is not yet possible. Accordingly, there is still a significant need for research in the development of potential therapeutic approaches and the investigation of causes.

Our focus is on the role of Artificial Intelligence in improving risk prediction and prevention. For example, AI could be employed in clustering to identify differentiable subgroups, such as previously unknown geno-/phenotype associations.