Data protection and deep learning: How Peiyao Mao makes medical image data securely usable

How can medical image data be analysed in such a way that it provides insights without revealing sensitive information about patients? Peiyao Mao from the Otto Friedrich University of Bamberg has found an innovative answer to this question: ‘My thesis deals with how modern vision transformer (ViT) models and special hashing methods can help to efficiently search image data while protecting privacy.’

By converting the image information into compact, hashed forms, the content remains semantically meaningful - without allowing conclusions to be drawn about personal data.

Between research and studies

Despite her success, Peiyao has not continued her research for the time being. ‘I'm currently concentrating on my Master's degree and would like to familiarise myself with many more modules.’ But the award has left a positive mark: New contacts, fresh motivation - and a clear focus on digital health. ‘I was able to exchange ideas with many interesting people from the field of medical informatics and engineering, which has motivated me to do more research in this area in my Master's programme.’

Peiyao became aware of the award through her supervisor - and through a fellow student who had already taken part in the DMEA sparks Award the previous year and also achieved third place in the ‘Bachelor's thesis’ category.

Tips for applicants

Peiyao's most important advice for students with DMEA sparks Award ambitions: Don't forget the practical relevance! ‘The research topic should not only be academically interesting, but also practically applicable.’ At the DMEA in particular, it is crucial that companies can work with the results - and that the exposé is scientifically convincing.