This position is part of a joint collaboration between the two largest research programs in Sweden, the Wallenberg AI, Autonomous Systems and Software Program (WASP) and the SciLifeLab and Wallenberg National Program for Data-Driven Life Science (DDLS), with the ultimate goal of solving ground-breaking research questions across disciplines.
Our Research
We are a translational group with a strong cross-disciplinary collaboration between breast radiology at Karolinska Institutet (KI) and computer science at the Royal School of Engineering (KTH) and SciLifeLab. The members at KTH are mainly responsible for the development of algorithms, and the members at KI are mainly responsible for directing the initial data curation and the later evaluation of algorithm performance. In addition, we are exploring aspects of AI governance, including setting up a national platform for retrospective validation of AI algorithms in mammography screening and to develop early-warning methods to detect when the AI predictions might no longer be trustworthy.
We now seek to strengthen our internationally leading clinical research group at KI with a postdoctoral researcher within the field of biostatistics applied to clinical studies of AI in breast cancer imaging.
Your mission
Your research will consist of mainly two different topics. The first topic concerns the development and evaluation of in-house AI algorithms. One of your responsibilities will be to contribute with statistical competence in the development process of AI algorithms, helping the computer scientists with appropriate evaluation methods and with study design. You will also take responsibility for the retrospective evaluation of the developed algorithms in our data and in data from international collaborators. The second topic concerns the real-world follow-up of the performance and trustworthiness of AI algorithms in general, both in-house and commercial ones. You would evaluate different approaches to detect when a “black-box” AI algorithm may no longer be trustworthy, such as detecting distributional shifts. You might also explore the use of 3D-printed phantoms as reference standards for daily quality control of installed AI algorithms.