01 Research
I build self-supervised and supervised AI that finds rare diseases and under-represented anatomies without labelled examples, working closely with clinicians to make it trustworthy enough for the clinic.
01 Statement
Medical-imaging AI has matured quickly under conditions that hold in a handful of large hospitals: abundant labelled data, a narrow set of imaging protocols, and dense clinician and engineer collaboration. Most of my work asks what happens when those conditions don't hold: when a finding is rare enough that no one has labelled an example of it, when a scanner or protocol differs from the one a model was trained on, or when a clinician needs to know how much to trust a prediction rather than just receiving one.
01
I build methods that find rare or unexpected findings in medical images without labelled abnormal examples, including confidence-aware self-supervised localisation and zero-shot pipelines built on pre-trained segmentation models.
02
A point prediction without a confidence signal has limited clinical use. I develop Laplace-approximation-based methods, including L-FUSION for fetal ultrasound segmentation, that give clinicians an interpretable measure of how much to trust a given output.
03
Some anatomy is chronically under-represented in training data. I build diffusion-based synthesis pipelines, for female pelvic MRI and separately for label-free cardiac ultrasound, that generate realistic training and benchmark data where real examples are scarce.
04
Small-cohort and low-resource settings can't always afford backpropagation-heavy training. I've explored alternatives including Forward-Forward networks.
05
Methods only matter if they hold up on real clinical questions. I collaborate directly with clinical groups on lymphatic perfusion in Fontan patients, pancreas segmentation in diabetes, and axillary lymph-node metastasis prediction, work that keeps the other four threads honest about what a hospital actually needs.
Two placements in international medical-imaging challenges (winner of the MOOD Challenge in 2023 and 2024 including 5-year GOAT prize, third place at the VAND Challenge in 2023) within amazing teams.
03 Workshops & proceedings
Organising and editing
2026
Co-Organiser
MICCAI CAPI & WOMEN Workshop (Computer-Aided Pelvic Imaging for Female Health)
2025
Co-Organiser & Editor
MICCAI 1st International Workshop on Computer-Aided Pelvic Imaging for Female Health (CAPI), Daejeon, South Korea
Edited proceedings, with the 10th International Workshop on Skin Image Analysis (ISIC 2025): Skin Image Analysis, and Computer-Aided Pelvic Imaging for Female Health, Springer Nature. Co-edited with M. E. Celebi, C. Barata, A. Halpern, P. Tschandl, M. Combalia, and others, including M. May.
2023
Co-Organiser & Editor
MICCAI 4th International Workshop on Advances in Simplifying Medical Ultrasound (ASMUS), Vancouver, Canada
Edited proceedings: Simplifying Medical Ultrasound, Springer Nature LNCS vol. 14337. Co-edited with B. Kainz, A. Noble, J. Schnabel, B. Khanal, and T. Day.
04 Reviewing
Reviewing for
05 Output
Selected papers
MICCAI · 2026
@inproceedings{muller2026fibers,
title = {Fibers of Asymmetric Similarity: A Framework for
Clinical and Imaging Data},
author = {M{\"u}ller, Johanna P. and Baugh, M. and Wright, R.
and Day, T. and Rezavi, R. and Kainz, B.},
booktitle = {MICCAI},
year = {2026}
}Simplifying Medical Ultrasound, ASMUS 2025, eds. Ni D, Noble A, Huang R, Xue W, Springer LNCS vol. 16165, Cham
@inproceedings{muller2026lfusion,
title = {L-FUSION: Laplacian Fetal Ultrasound Segmentation and
Uncertainty Estimation},
author = {M{\"u}ller, Johanna P. and Wright, Robert and Day, Thomas G. and Venturini, Lorenzo and Budd, Samuel F. and Reynaud, Hadrien and Hajnal, Joseph V. and Razavi, Reza and Kainz, Bernhard},
editor = {Ni, D. and Noble, A. and Huang, R. and Xue, W.},
booktitle = {Simplifying Medical Ultrasound},
series = {Springer LNCS},
volume = {16165},
address = {Cham},
year = {2026},
note = {ASMUS 2025}
}MICCAI CAPI Workshop · 2025
@inproceedings{muller2026diffusing,
title = {Diffusing the Blind Spot: Uterine MRI Synthesis with
Diffusion Models},
author = {M{\"u}ller, Johanna P. and Knupfer, A. and Bl{\"o}ss, P.
and Vittur, E. B. and Kainz, B. and Hutter, J.},
editor = {Celebi, M. E. and others},
booktitle = {Skin Image Analysis, and Computer-Aided Pelvic Imaging
for Female Health},
series = {Springer LNCS},
volume = {16149},
address = {Cham},
year = {2025},
note = {CAPI 2025}
}Proceedings of the AAAI Conference on Artificial Intelligence, 38(19):21037–21045 · 2024
@inproceedings{dombrowski2024tradeoffs,
title = {Trade-offs in Fine-Tuned Diffusion Models Between
Accuracy and Interpretability},
author = {Dombrowski, M. and Reynaud, H. and M{\"u}ller,
Johanna P. and Baugh, M. and Kainz, B.},
booktitle = {Proceedings of the AAAI Conference on Artificial
Intelligence},
volume = {38},
number = {19},
pages = {21037--21045},
year = {2024},
month = mar
}International Workshop on Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, pp. 177–187, Cham: Springer Nature Switzerland · 2023
@inproceedings{muller2023confidenceaware,
title = {Confidence-Aware and Self-supervised Image Anomaly
Localisation},
author = {M{\"u}ller, Johanna P. and Baugh, M. and Tan, J.
and Dombrowski, M. and Kainz, B.},
booktitle = {International Workshop on Uncertainty for Safe
Utilization of Machine Learning in Medical Imaging},
publisher = {Springer Nature Switzerland},
address = {Cham},
pages = {177--187},
year = {2023},
month = oct
}International Conference on Medical Image Computing and Computer-Assisted Intervention, pp. 162–172, Cham: Springer Nature Switzerland · 2023
@inproceedings{baugh2023many,
title = {Many Tasks Make Light Work: Learning to Localise
Medical Anomalies from Multiple Synthetic Tasks},
author = {Baugh, M. and Tan, J. and M{\"u}ller, Johanna P.
and Dombrowski, M. and Batten, J. and Kainz, B.},
booktitle = {International Conference on Medical Image Computing
and Computer-Assisted Intervention},
publisher = {Springer Nature Switzerland},
address = {Cham},
pages = {162--172},
year = {2023},
month = oct
}