Visualizing Propagator-Based Diffusion Imaging Data

• Applicant: Professor Dr.-Ing. Thomas Schultz
• Subject Area: Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
• Term: from 2019 to 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 422414649

• Final Report Year: 2024

Final Report Abstract

Diffusion-weighted magnetic resonance imaging is firmly established for imaging the white matter of the brain both in research and in clinical settings. More and more frequently, diffusion is measured not only with high angular resolution, but also with different diffusion weightings. The diffusion propagators estimated from such data contain more detailed information about tissue microstructure. In this project, we have established new approaches for visualizing these complex data, taking into consideration suitable representations, the reduction of various types of uncertainty, and the comparative visualization of multiple datasets. First, we developed a system for volume rendering of these data. We learned a suitable dimensionality-reduced representation in a data-driven manner using a deep neural network architecture that can be trained in an unsupervised fashion. The structure of the architecture was designed to ensure that both local and regional features are available when interactively learning a transfer function with a Random Forest. Second, we developed approaches to reduce different types of uncertainties that affect visualizations based on common tractography methods. We found that model uncertainty, which previously received little attention, has relevant impacts on the completeness of tractography results, and that it can be effectively reduced by weighted averaging of the results from multiple models. Additionally, we developed two new approaches to reduce data uncertainty arising from measurement noise: one through the joint approximation of fODF tensors in a spatial neighborhood, the other through combining a tensor-based representation of the data with an Unscented Kalman Filter (UKF). Both approaches were successfully evaluated on clinical datasets from our collaboration partner. Finally, we extended the UKF-based approach to consider not only crossing, but also fanning fiber bundles, which led to a more complete reconstruction of corresponding tracts. Our third goal was develop a comparative visualization that makes it possible to extract diagnostically relevant information from these large and complex datasets in a time-efficient manner. As a first step towards that goal, we developed a system that mathematically models the variability occurring in a healthy control group, so that it can guide the user towards deviations that indicate potentially problematic abnormalities. For a more detailed analysis, we designed a so-called anomaly lens. To permit a broad application on existing clinical data, this system still uses a traditional diffusion tensor model.

Publications

• “Interactive Classification of Multi-Shell Diffusion MRI With Features From a Dual-Branch CNN Autoencoder.” Proc. EG Workshop on Visual Computing for Biology and Medicine (VCBM) 1-11, 2020. Honorable Mention Award.
  A. Torayev & T. Schultz
  
• “Reducing Model Uncertainty in Crossing Fiber Tractography.” Proc. EG Workshop on Visual Computing for Biology and Medicine (VCBM) 55-64, 2021. Best Paper Award.
  J. Gruen, G. van der Voort & T. Schultz
  
• Model Averaging and Bootstrap Consensus‐based Uncertainty Reduction in Diffusion MRI Tractography. Computer Graphics Forum, 42(1), 217-230.
  Gruen, J.; van der Voort, G. & Schultz, T.
  
• Parcellation-Free prediction of task fMRI activations from dMRI tractography. Medical Image Analysis, 76, 102317.
  Khatami, Mohammad; Wehler, Regina & Schultz, Thomas
  
• Anisotropic Fanning Aware Low-Rank Tensor Approximation Based Tractography. Lecture Notes in Computer Science, 140-151. Springer Nature Switzerland.
  Gruen, Johannes; Sieg, Jonah & Schultz, Thomas
  
• Spatially regularized low-rank tensor approximation for accurate and fast tractography. NeuroImage, 271, 120004.
  Gruen, Johannes; Groeschel, Samuel & Schultz, Thomas
  
• “Detection and Visual Analysis of Pathological Abnormalities in Diffusion Tensor Imaging with an Anomaly Lens.” In: Proc. EuroVis Short Papers, 43–47, 2023.
  M. Bareth, S. Groeschel, J. Gruen, P. Pretzel & T. Schultz