About Me
I am a PhD student under the supervision of Prof. Dr. Klaus Maier-Hein at the German Cancer Research Center (DKFZ), Germany. I am also affiliated with Heidelberg University, Germany.
My research interests broadly intersect machine learning and medical image analysis.
My PhD thesis focuses on representation learning in Transformers and self-attention based networks for segmenting organs and pathologies in radiological images. The insights from my work demonstrate noticeable challenges for transformer-based networks for learning representations for on small and sparsely-annotated medical image datasets. My work also demonstrates that convolutional architectures with Transformer-based scaling are ideal for medical image segmentation.
Updates
- 2024-07: 5 months internship at Siemens Healthineers, Princeton, NJ, USA.
- 2024-07: 1 paper accepted at ECCV 2024! (first author)
- 2024-06: 1 paper accepted in MICCAI 2024!
- 2023-10: Short research stay at Imaging AI lab at Luxembourg Centre for Systems Biomedicine, University of Luxembourg, Luxembourg.
- 2023-06: Invited talk at Medical and Environmental Computing Lab at TU Darmstadt, Germany.
- 2023-06: 1 short paper accepted at MIDL 2023! (first author)
- 2023-05: 2 early accepts at MICCAI 2023 (top 14% of over 2k submissions), including 1 first author paper!
- 2022-11: 1 poster accepted at MedNeurIPS 2022! (first author)
- 2021-03: PhD position at Division of Medical Image Computing (MIC) at the German Cancer Research Center (DKFZ), Germany under Prof. Dr. Klaus H. Maier-Hein.
- 2021-01: Graduated from the University of Bonn, Germany (M.Sc., Computer Sc.).
- 2021-01: Defended Masters thesis on "Fully 3D Segmentation of Neuroanatomy", completed at the German Center for Neurodegenerative Diseases (DZNE), Germany under Prof. Dr. Martin Reuter.
Major Publications
MedNeXt: Transformer-driven Scaling of ConvNets for Medical Image Segmentation
First author, MICCAI 2023 (Core Ranking: A)
Early Accept, Top 14% of submissions
MedNeXt is a fully ConvNeXt architecture designed for state-of-the-art 3D medical image segmentation. It beats a wide range of ConvNet and Transformer based baselines on a number of challenging public datasets.
Code: https://github.com/MIC-DKFZ/MedNeXt Paper: https://arxiv.org/abs/2303.09975
