Computer scientists at FAU develop a tool for automatically analyzing results of barium swallow tests, and receive 390,000 euros in funding for their project.
Using AI to analyze the swallowing process, especially in elderly individuals and Parkinson’s patients, can make assessments far more objective and precise. Biomedical computer scientists at Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) have developed a tool that automatically analyzes the results of barium swallow tests and detects any pathological and potentially life-threatening changes to the swallowing mechanism. The German Research Foundation (DFG) is providing approximately 390,000 euros in funding for the project.
Swallowing is an automatic reflex that we scarcely pay attention to. For elderly people, after a stroke or in the case of neurological diseases, however, the physiology of this process is often disrupted. “If pieces of food remain in the throat or find their way into the respiratory tract, this can lead to serious complications and even death,” explains Prof. Dr. Andreas Kist, specialist for biomedical signal analysis at FAU. Statistics on Parkinson’s disease highlight the medical importance of this issue: pneumonia caused by inhaling food is the most frequent cause of death for this group of patients, accounting for 70 percent of all deaths.
Neural networks learn anatomy
One established method for investigating the physiology of swallowing are videos of barium swallow tests. They track the bolus, the scientific name for the mass of chewed food, as it progresses down the food pipe to the stomach. A resolution of 30 images per second delivers the best results. However, evaluating the images is a time-consuming manual task, by its very nature subjective and prone to error.
The Kist Group is therefore working on an AI-assisted procedure that automatically analyzes videos. This is certainly not a trivial task, as swallowing is a closely-coordinated process involving various muscles, nerves, bones and cartilage. At the end of the day, the tool must be able to identify the anatomical landmarks in each individual patient, even if the images have been taken from different perspectives and using devices from different manufacturers.
“In the first project phase, our priority is to train the neural networks with data on anatomy,” explains Luisa Neubig. “The hyoid bone, larynx, esophagus, windpipe – AI must be able to recognize all of these body parts accurately before it can come to any conclusions regarding the physiology of the swallowing process.” Neubig is a doctoral candidate at Andreas Kist’s Chair and will lead the project that is due to run for three years. Luisa Neubig, who is originally from Nuremberg, studied Medical Engineering at FAU and already explored deep learning models for analyzing swallowing in her Master’s thesis. In 2023, she won the DMEA Newcomers Award for her Master’s thesis, awarded by the DMEA, the exhibition for digital health and applications.
Bologram is hoped to accelerate decision-making process
In the second phase of the project, the aim is to make the model able to use the anatomical parameters it has learned to follow the progress of the food and to recognize if any residue is left behind. In the third phase, the researchers will focus on transferring all data on the swallowing process into a standardized grid and present it in a compressed image known as a bologram. “The bologram should allow decisions to be taken quickly in clinical practice,” explains Luisa Neubig. “We also work with dyes that indicate at a glance whether everything is as it should be, or whether intervention is required.”
The work of the biomedical computer scientists from Erlangen may contribute towards making barium swallow tests more widespread across the German health system. This procedure is commonplace in the USA, but in Europe doctors tend to prefer endoscopic examinations to avoid the radiation inevitably entailed by barium swallow tests,” explains Andreas Kist. “However, endoscopies do not show the actual act of swallowing and are focused instead on a comparison of before and after.” Ideally, the reliability of the new AI-assisted tool will allow barium swallowing test videos to require fewer images per second or to be required less frequently during treatment, both of which would lead to a considerable reduction in the doses of radiation patients are exposed to.
An innovation hub for artificial intelligence in medicine
The DFG is providing approximately 390,000 euros in funding. The funding will be used to finance a doctoral research position for a period of three years. This funding is further proof of FAU’s special expertise as a location for innovation and a hub for artificial intelligence in medicine. The professorship led by Andreas Kist has been established at the Department of Artificial Intelligence in Biomedical Engineering (AIBE). The AIBE was established as part of the High-Tech Agenda Bavaria in late 2019 and takes an interdisciplinary and cross-subject approach at the intersection of medicine and engineering.
Further information:
Prof. Dr. Andreas Kist
Professorship for Artificial Intelligence in Communication Disorders
andreas.kist@fau.de