Julian McGinnis

I'm a PhD student and researcher in Computer Science at the Munich Centre for Machine Learning (MCML), hosted at the Technical University of Munich (TUM) and TUM Klinikum Rechts der Isar. I work at the intersection of medical imaging, neural fields, and reconstruction algorithms. My research focuses on implicit neural representations with applications to neural compression and super-resolution. I also develop tools for MS lesion segmentation in brain and spinal-cord MRI for Multiple Sclerosis.

Before my PhD, I studied electrical engineering at the Duale Hochschule Baden-Württemberg (DHBW) (B.Eng.) and Technical University of Munich (TUM) (M.Sc.), and worked for several years as an engineer in embedded systems and security for the RAFI Group , designing PCBs and programming microcontrollers in C/C++ (and a sprinkle of ASM), with a particular focus on human–machine interfaces and capacitive sensing.

Email  /  Scholar  /  Twitter  /  Github

Research (Selected Publications)

I'm interested in machine learning, neural compression and signal modeling. Some papers are highlighted.

We introduce a scalable meta-learning framework tailored to learning medical neural fields, and provide some theoretical insights showing how SIREN scales learning rates.

We propose a robust deep learning framework for the automatic segmentation of spinal cord and MS lesions on axial T2-weighted MRI scans.

We introduce SINR - a method to parameterize the continuous deformable transformation represented by an INR using Free Form Deformations (FFD) enabling robust multi-modal registration while preventing spatial folding.

We present LST-AI, an open-source deep learning ensemble for multiple sclerosis lesion segmentation that specifically addresses class imbalance and significantly outperforms traditional tools like LST and standard U-Net architectures.

By modeling complementary MRI contrasts in a shared anatomical space with INRs, we are able to super-resolve MRI scans using single subject data alone.

We propose a novel family of segmentation models that learn a mapping from a real-valued coordinate space to a shape representation, enabling the segmentation of anatomical shapes in high-dimensional continuous spaces even with sparse or partial data.

We present a biological dataset of whole-brain vessel graphs, now part of Stanford's Open Graph Benchmark Dataset.

MICCAI 2024 Tutorial on Implicit Neural Representations for Medical Imaging

Julian McGinnis, Nil Stolt-Ansó, Veronika Spieker, Suprosanna Shit, Maik Dannecker, Vasiliki Sideri-Lampretsa, Björn Menze, Daniel Rückert, Benedikt Wiestler

Based on Jon Barron's website.