Thesis Demographic Bias in Histopathology

Fraunhofer MEVIS is a world leading and internationally connencted research center for computer assistance in medicine. With about 140 employees, our mission is to conduct patient-centric research and development to improve clinical processes for the benefit of our clinical partners and, in the end, patients.

What you will do

It has been observed that deep learning models are able to identify patient characteristics such as age, sex, and self-reported race with high accuracy from medical images such as chest x-ray recordings, even when medical doctors cannot. This raises the potential for such models to learn to (falsely) diagnose patients of different demographics differently, even if they present with the same disease characteristics. This may happen because, for example, some groups may tend to be misdiagnosed or underdiagnosed in the datasets using which such models are trained, or simply because these groups happen to be (spuriously) correlated with the outcome variable in the training dataset. The model would then (undesirably) learn to rely on such spurious correlations, resulting in poor model robustness. The risk of such demographic shortcut learning in histopathological images has not been investigated until now; filling this gap is the aim of this thesis.   The topic is suitable for either a B.Sc. thesis or a M.Sc. thesis. 

Yor tasks to accomplish may include the following: 

• Investigating how well demographic patient attributes and other potential shortcut features can be predicted from histopathological images alone.  

• Investigating how well outcomes can be predicted purely based on these potential shortcut features, and whether current deep learning methods perform significantly better than this.  

• Investigating whether current deep learning models in histopathology suffer from (demographic) shortcut learning, and if yes, how this can be mitigated.

What you bring to the table

• Enrolled student of computer science or similar field of study

• Prior knowledge of and first experiences with deep learning  

• Knowledge of python 

• An interest in medical image analysis / histopathology and trustworthy AI. (Prior knowledge or experience in these areas are a plus but not required.) 

What you can expect

• a friendly working environment at one of our sites in Bremen, Lübeck, or Hannover 
• self-determined work and the freedom to create new tasks 
• work within an interdisciplinary team of software developers and image processing researchers 
• ideal conditions for practical experience alongside your studies 
• supervision by experienced researchers in histopathology AI and AI fairness for medical images 

The position is limited until the completion of the thesis.

We value and promote the diversity of our employees' skills and therefore welcome all applications - regardless of age, gender, nationality, ethnic and social origin, religion, ideology, disability, sexual orientation and identity. Severely disabled persons are given preference in the event of equal suitability. 

With its focus on developing key technologies that are vital for the future and enabling the commercial utilization of this work by business and industry, Fraunhofer plays a central role in the innovation process. As a pioneer and catalyst for groundbreaking developments and scientific excellence, Fraunhofer helps shape society now and in the future. 

Interested? Apply online now. We look forward to getting to know you! Applications are possible in German or English. Please include a cover letter, your CV, your last transcript of records and your envisioned starting date of the thesis.

If you have any work related questions regarding this position, do not hesitate to contact: 

Dr. Johannes Lotz (johannes.lotz@mevis.fraunhofer.de) 

Dr. Eike Petersen (eike.petersen@mevis.fraunhofer.de) 

Fraunhofer Institute for Digital Medicine MEVIS 

www.mevis.fraunhofer.de 

Requisition Number: 78346                Application Deadline: