Fraunhofer Institute for Industrial Mathematics ITWMMedicine, technology and science: her studies in Nanjing, China, Gothenburg, Sweden, and Munich, Germany, already paved the way for Dr. Xiaoyin Cheng. She wanted to do research and do good. This is exactly what she does today at the Fraunhofer Institute for Industrial Mathematics ITWM in Kaiserslautern.
Puzzles Are the Best Medicine: Xiaoyin Wants to Save Lives with Pictures
This Article Is Taken from Our Book »Forscherinnen im Fokus – Wir schaffen Veränderung« (»Female Researchers in Focus – We Create Change«)
At first glance, »image processing« doesn't make you think of lifesavers. But if you take a closer look, you realize that it can become one! Xiaoyin, for example, is working intensively on the medical reconstruction of 4D images or the optimization of positron emission tomography (PET). Both are used in the early detection and treatment of cancer, among other things.
It Never Gets Boring
Self-employment, a career in business? Xiaoyin hasn't even thought about it yet. She is convinced that she has found her place in research. She has been at Fraunhofer for more than eight years now, developing algorithms for image processing, programming, training and discussing with her colleagues until a solution to a specific problem seems to have been found. »Research is a bit like doing a puzzle, it never gets boring. For me, that's its biggest advantage over all other jobs. Every project is different, you have to face a new challenge every time.«
Long Breath...
Sometimes such a puzzle has more than 1,000 pieces. Then you need a lot of patience. It helps all the more if you can openly exchange ideas within the team and put the pieces together. »The roles in a research project are clearly assigned. For me, it's great that project managers give me time to implement exciting ideas and have my back by taking over communication with project partners. This allows me to concentrate fully on my research work.«
... And Passionate Learning
For Xiaoyin, science has another major advantage: you learn – all the time. »You learn as you live«, she says, meaning not the dry 'learning for life' often cited by teachers, but learning from your own motivation, from your own drive. Something new is constantly developing in the field of artificial intelligence. »You have to stay up to date.« This is another reason why research is just right for Xiaoyin, as it allows her to deal with new topics every day – quantum computing is one of them! She is constantly expanding her extensive knowledge of this innovative technology and, as co-head of the Rhineland-Palatinate Quantum Initiative »QUIP«, ensures that young researchers are offered a detailed research program for training, further education and networking.
True Crime Instead of an End in Itself: AI and Algorithms for Fraud Detection
According to police statistics, there were 801,412 fraud offenses in Germany in 2022. This includes billing fraud in the care sector, which causes enormous financial damage to both private individuals and insurance companies. By developing AI software, Xiaoyin and her colleagues from the Fraunhofer ITWM's Image Processing department want to make it easier for the authorities to prosecute cases. They are conducting research on the »PflegeForensik« project together with project manager Dr. Henrike Stephani, the Dresden Public Prosecutor General's Office and the Leipzig Police Department's Economic Crime Unit.
Care services rarely work in a standardized or even fully automated way. Instead, there are many different types of documents, lots of paper, many written notes, abbreviations instead of signatures. If there is suspicion of fraud, investigating authorities have to laboriously transfer all the documents into Excel sheets, juxtapose many small individual records and compare them – previously by hand! Because this eats up time and resources, the analysis is often aborted as soon as the burden of proof is sufficient for a guilty or acquittal verdict.
Data volumes are important for any AI: in the project, the team uses automated image processing with machine learning to recognize important document types such as route planning, visit reports, performance records or duty rosters and record them digitally in a central database.
Care services rarely work in a standardized or even fully automated way. Instead, there are many different types of documents, lots of paper, many written notes, abbreviations instead of signatures. If there is suspicion of fraud, investigating authorities have to laboriously transfer all the documents into Excel sheets, juxtapose many small individual records and compare them – previously by hand! Because this eats up time and resources, the analysis is often aborted as soon as the burden of proof is sufficient for a guilty or acquittal verdict.
Machine Learning for Data Acquisition
Things will soon be simpler and more up-to-date with the AI software developed by the researchers from the »PflegeForensik« project. On the one hand, they are working closely with the police to tailor the solution precisely to their needs and thus create a real advantage for everyday life. On the other hand, they are using automated image processing with machine learning to recognize important document types such as tour planning, visit reports, performance records or duty rosters and record them digitally in a central database. Image processing and modern deep learning methods are combined for this purpose.
Proof by Algorithm
An intelligent algorithm is also programmed to detect anomalies. For example, does the tour planning match a registered service? Can the visit report be correct if the nurse mentioned there was not actually on duty? In order to process this information and automatically detect anomalies, the algorithm is first fed with thousands of artificial and anonymized data records, trained and continuously improved with data from real cases.
The AI Does Not Condemn
In addition to the development and ongoing training of the algorithm, the scientists have to deal with other challenges – be it the ease of use of the software solution or the computing power required to process countless data sets. They also have to comply with the law so that the investigation results stand up in court. One thing must also be clear at all times: AI does not convict! It supports efficient investigations, but human verification will remain necessary in the future.