Mélanie Roschewitz (née Bernhardt)

PhD student @ Imperial College London - Safe Machine Learning for Medical Imaging.

resized_cv.jpg

About me

I’m a 3rd year Computing PhD student in the BiomedIA lab at Imperial College London, supervised by Prof. Ben Glocker and fully funded by an Imperial College President’s scholarship. My research interests cover various aspects of medical imaging machine learning, in particular reliability and safety of these systems.

Background

I started my studies with a B.Sc. in Mathematics as well as a postgraduate diploma in Statistics from the University of Strasbourg (France). Later, I graduated from ETH Zurich with a M.Sc. in Data Science, focusing on medical applications of machine learning. I then joined Microsoft Research Cambridge (UK), where I was an Applied Research in the InnerEye team, focusing on developping ML models for personalized treatment as well as on open-source efforts in ML for healthcare. In October 2021, I then started my PhD at Imperial. I was also a intern in the ML research team of Kheiron Medical Technologies from April to October 2022.

News

Oct 19, 2023 Our work Automatic correction of performance drift under acquisition shift in medical image classification has been published in Nature Communications
Oct 08, 2023 Looking forward to present our work Distance Matters For Improving Performance Estimation Under Covariate Shift at the Uncertainty Quantification Workshop at ICCV 2023.
Mar 04, 2023 Excited to present our latest results on automatic correction of performance drift as an oral at ECR, the European Congress of Radiology in Vienna.
Oct 01, 2022 Our paper Failure Detection in Medical Image Classification: A Reality Check and Benchmarking Testbed has been published in Transactions on Machine Learning Research.
Jun 12, 2022 Our comment Potential sources of dataset bias complicate investigation of underdiagnosis by machine learning algorithms has been published in Nature Medicine.

Selected publications

  1. Automatic correction of performance drift under acquisition shift in medical image classification
    Mélanie Roschewitz ,  Galvin Khara ,  Joe Yearsley , and 7 more authors
    Nature Communications, Oct 2023
  2. Active label cleaning for improved dataset quality under resource constraints
    Mélanie Bernhardt ,  Daniel C Castro ,  Ryutaro Tanno , and 8 more authors
    Nature communications, Mar 2022
  3. Potential sources of dataset bias complicate investigation of underdiagnosis by machine learning algorithms
    Mélanie Bernhardt ,  Charles Jones ,  and  Ben Glocker
    Nature Medicine, Mar 2022
  4. Failure Detection in Medical Image Classification: A Reality Check and Benchmarking Testbed
    Mélanie Bernhardt ,  Fabio De Sousa Ribeiro ,  and  Ben Glocker
    Transactions on Machine Learning Research, Mar 2022