Saikat Roy
Senior Machine Learning Engineer @ mediaire GmbH. Berlin, Germany.
contact@saikatroy.me
I am a Senior Machine Learning Engineer at mediaire GmbH, leading AI development for MRI-based medical image analysis products, and a Visiting Researcher at the Diagnostic Image Analysis Group (DIAG) at Radboud University Medical Center, working on AI for prostate cancer diagnosis.
I was previously a Doctoral Researcher at the German Cancer Research Center (DKFZ) and Heidelberg University, where my PhD research focused on Transformers and Self-Attention for Medical Image Segmentation. One of the key contributions of my Ph.D. research is the development of MedNeXt, a 3D ConvNeXt-based architecture for medical image segmentation, which has been widely adopted (700+ citations) in the research community and has ranked first in multiple independent benchmark evaluations (NeurIPS 2024, MICCAI 2024), as an improvement upon the popular nnUNet.
During my Ph.D. I also spent time as a Research Intern at Siemens Healthineers, USA developing patch-based diffusion models for medical image denoising and super-resolution. Before my Ph.D., I completed an M.Sc. in Computer Science at the University of Bonn, working alongside the German Center for Neurodegenerative Diseases (DZNE) on 3D deep representation learning for neuroanatomy segmentation.
My research specializes in predictive and generative problems in 3D medical image analysis, particularly focusing on effective large-scale 3D representation learning.
news
| Aug 14, 2026 | My PhD work on MedNeXt (MICCAI 2023) was selected as a finalist for the MICCAI 2026 Young Scientist Publication Impact Award. |
|---|---|
| Apr 25, 2026 | Presented on large-scale supervised representation learning with ConvNeXts in Medical Image Segmentation at the Sprint AI Training for African Medical Imaging Knowledge Translation (SPARK) Academy. |
| Mar 26, 2026 | Gave a talk on large-scale supervised representation learning with MedNeXt-v2 at Raysearch Laboratories, Sweden. |
| Jul 08, 2025 | Spoke on Trend-Chasing in Medical AI at the Machine Learning Galore Talks series at Heidelberg University, Germany. |
| Jan 02, 2025 | Research visit to the Diagnostic Image Analysis Group (DIAG) at Radboud University Medical Center, Netherlands. |
| Oct 10, 2024 | Delivered the keynote for the MICCAI 2024 Challenge Body Maps: Towards a 3D Atlas of the Human Body. |
| Oct 10, 2023 | Research visit and invited talk on Current Roadblocks for Transformers in Medical Image Segmentation at the Luxembourg Centre for Systems Biomedicine, University of Luxembourg, Luxembourg. |
| Jul 12, 2023 | Invited talk on Transformers for Medical Image Segmentation: Where do we stand? at the Medical and Environmental Computing Lab, TU Darmstadt, Germany. |
| Jul 02, 2023 | Presented on Transformers and Large Kernel Nets in Medical Image Analysis as part of Advanced Deep Learning Tutorials at BVM, Germany. |
| Jun 26, 2022 | Presented on Transformers in Medical Image Analysis as part of Advanced Deep Learning Tutorials at BVM, Germany. |
selected publications
- MICCAIMedNeXt: Transformer-driven scaling of convnets for medical image segmentationIn MICCAI 2023, 2023
- ECCVSkeleton Recall Loss for Connectivity Conserving and Resource Efficient Segmentation of Thin Tubular StructuresECCV 2024, 2024
- MICCAInnU-Net Revisited: A Call for Rigorous Validation in 3D Medical Image SegmentationMICCAI 2024, 2024
- NeurIPSTouchstone Benchmark: Are We on the Right Way for Evaluating AI Algorithms for Medical Segmentation?NeurIPS 2024, 2024
- ISBIInvestigating the Feasibility of Patch-Based Inference for Generalized Diffusion Priors in Inverse Problems for Medical ImagesIn IEEE International Symposium on Biomedical Imaging (ISBI), 2025
- TMLRPrimus: Enforcing Attention Usage for 3D Medical Image SegmentationTransactions on Machine Learning Research (TMLR), 2026