Max Pietsch

King's College London

I use machine learning to extract information from medical imaging data.

In the weeks around birth, a baby’s brain tissue undergoes drastic changes visible on routinely acquired grayscale magnetic resonance imaging scans. However, changes are spatially and temporally varied and it is challenging to extract information that is specific to the processes that are happening on a cellular and morphological level.

I develop methods to learn features from 6D medical imaging data (multi-shell high-angular resolution diffusion-weighted magnetic resonance imaging). These data capture the movement of water molecules in and between brain cells which allows probing orientation-resolved microscopical properties of the developing brain. I use data-driven methods to extract features that allow decomposing the images into interpretable components that are related to brain development. Using these, one can answer questions about normal and abnormal tissue maturation in terms of maturation patterns. With the associated orientation information, one can trace these components throughout white matter pathways of the brain and investigate brain maturation as an interconnected network over time. Applications of these tools facilitate the in-vivo study of normal development and the detection of brain-related abnormalities such as those associated with premature birth.

Data were released in the form of a normative atlas of neonatal brain development from data acquired as part of the developing Human Connectome Project. Images from this work were featured in news articles in Wired, BBC, Guardian, and Nature Outlook and are shown below: Neonatal tissue matruation contrast

Another line of my work uses deep learning to automatically detect and remove signatures of deteriorated data quality in diffusion MRI data. We train a convolutional neural network to make the data more consistent — without having access to non-corrupted data. Instead of supervised learning, we train the network using techniques based in statistical learning theory. Deep neural networks are powerful models but they can introduce artefacts into the images which is at odds with the need of medical data to be interpretable. We approach this problem by constraining the network to produce limited and well-defined changes that bound the scope of introducing local artefacts, therefore increasing the trust in the modified data. The code and pretrained neural network can be downloaded as Docker image from dStripe

In recent work on placental health modelling, we predict health scores using a fast and clinical T2* weighted MRI scan. This work utilises placental mean T2* as an age-dependent biomarker for placental health. The method is clinically appealing as it is fast and fully automated and it operates in two stages with human-interpretable output: deep-learning-based organ segmentation and normative Bayesian modelling similar to growth charts. In our clinical cohort with MRI acquired at 3T, it can be used to identify early placental insufficiency with an AUC of 0.95. This work was also featured in the press in King’s College London Spotlight on Impact (9 Aug 2021), Healthcare in Europe (30 Sep 2021), the Imaging Wire (16 Aug 2021), and European Hospital (Aug/Sept 2021). Placental health prediction

selected publications

For a more complete list, please see my google scholar.
  1. MEDIA
    dStripe: slice artefact correction in diffusion MRI via constrained neural network
    Pietsch, Maximilian, Christiaens, Daan, Hajnal, Joseph V, and Tournier, J-Donald
    Medical Image Analysis 2021
  2. MEDIA
    APPLAUSE: Automatic Prediction of PLAcental health via U-net Segmentation and statistical Evaluation
    Pietsch, Maximilian, Ho, Alison, Bardanzellu, Alessia, Zeidan, Aya Mutaz Ahmad, Chappell, Lucy C, Hajnal, Joseph V, Rutherford, Mary, and Hutter, Jana
    Medical Image Analysis 2021
  3. ISMRM
    3t++ temporally consistent 3 tissue HARDI decomposition of neonatal brain tissue
    Pietsch, Maximilian, Christiaens, Daan, Hutter, Jana, Cordero-Grande, Lucilio, Price, Anthony N, Hughes, Emer, Edwards, A David, Hajnal, Joseph V, Counsell, Serena J, and Tournier, J-Donald
    In 2021
  4. ISMRM
    Multi-component atlas of fetal brain development via decomposition of diffusion MRI
    Pietsch, Maximilian, Christiaens, Daan, Hutter, Jana, Cordero-Grande, Lucilio, Price, Anthony N, Hughes, Emer, Edwards, A David, Hajnal, Joseph V, Counsell, Serena J, and Tournier, J-Donald
    In 2020
  5. Neuroimage
    Cross-scanner and cross-protocol multi-shell diffusion MRI data harmonization: algorithms and results
    Ning, Lipeng, Bonet-Carne, Elisenda, Grussu, Francesco, Sepehrband, Farshid, Kaden, Enrico, Veraart, Jelle, Blumberg, Stefano B, Khoo, Can Son, Palombo, Marco, Coll-Font, Jaume, Scherrer, Benoit, Warfield, Simon K., Karayumak, Suheyla Cetin, Rathi, Yogesh, Koppers, Simon, Weninger, Leon, Ebert, Julia, Merhof, Dorit, Moyer, Daniel, Pietsch, Maximilian, Christiaens, Daan, Teixeira, Rui, Tournier, Jacques-Donald, Zhylka, Andrey, Pluim, Josien, Parker, Greg, Rudrapatna, Umesh, Evans, John, Charron, Cyril, Jones, Derek K., and Tax, Chantal W. M.
    Neuroimage 2020
  6. Neuroimage
    A framework for multi-component analysis of diffusion MRI data over the neonatal period
    Pietsch, Maximilian, Christiaens, Daan, Hutter, Jana, Cordero-Grande, Lucilio, Price, Anthony N, Hughes, Emer, Edwards, A David, Hajnal, Joseph V, Counsell, Serena J, and Tournier, J-Donald
    Neuroimage 2019
  7. Neuroimage
    MRtrix3: A fast, flexible and open software framework for medical image processing and visualisation
    Tournier, J-Donald, Smith, Robert, Raffelt, David, Tabbara, Rami, Dhollander, Thijs, Pietsch, Maximilian, Christiaens, Daan, Jeurissen, Ben, Yeh, Chun-Hung, and Connelly, Alan
    Neuroimage 2019
  8. ISMRM
    Transfer learning and convolutional neural net fusion for motion artefact detection
    Kelly, Christopher, Pietsch, Maximilian, Counsell, Serena, and Tournier, J-Donald
    In 2017
  9. World J Cardiol
    Optimal C-arm angulation during transcatheter aortic valve replacement: Accuracy of a rotational C-arm computed tomography based three dimensional heart model
    Veulemans, Verena, Mollus, Sabine, Saalbach, Axel, Pietsch, Maximilian, Hellhammer, Katharina, Zeus, Tobias, Westenfeld, Ralf, Weese, Jürgen, Kelm, Malte, and Balzer, Jan
    World J Cardiol 2016