Autoencoder-based anomaly mapping and visualization of white matter deformation with application to glioma cases

Introduction: Gliomas can cause structural deformation in white matter pathways, impacting both function and diagnosis. Currently, there is no method to map these alterations across the entire brain for treatment planning purposes. This study proposes a novel framework that utilizes a convolutional autoencoder to learn geometry-preserving streamline embeddings for mapping structural connectivity anomalies.

Methods: The autoencoder was trained and validated on 100 million synthetic streamlines. The anomaly-mapping pipeline was evaluated using data from 118 healthy subjects. The researchers analyzed how different tractography algorithms influenced anomaly scores and demonstrated the feasibility of the framework in five glioma patients using multi-shell and single-shell diffusion MRI scans.

Results: The study found that latent-space distances strongly correlated with native-space geometric distances (R2 = 0.82-0.98), indicating that the learned embeddings preserved streamline geometry. Anomaly scores varied depending on both the input and reference tractography methods. Patient anomaly maps revealed elevated scores near gliomas, as well as extending distally along affected pathways. These whole-brain anomaly maps were generated within minutes.

Conclusion: This framework provides a practical means to quantify and visualize geometric deviations in whole-brain tractograms, offering novel insights beyond traditional network and microstructural analyses. The findings have implications for clinical applications where white matter deformation is relevant.