Zero-shot fMRI decoding with three-dimensional registration based on diffusion tensor imaging

Takuya Fuchigami, Yumi Shikauchi, Ken Nakae, Manabu Shikauchi, Takeshi Ogawa, Shin IshiiScientific 2018 · DOI

Abstract

Functional magnetic resonance imaging (fMRI) acquisitions include a great deal of individual variability. This individuality often generates obstacles to the efficient use of databanks from multiple subjects. Although recent studies have suggested that inter-regional connectivity reflects individuality, conventional three-dimensional (3D) registration methods that calibrate inter-subject variability are based on anatomical information about the gray matter shape (e.g., T1-weighted). Here, we present a new registration method focusing more on the white matter structure, which is directly related to the connectivity in the brain, and apply it to subject-transfer brain decoding. Our registration method based on diffusion tensor imaging (DTI) transferred functional maps of each individual to a common anatomical space, where a decoding analysis of multi-voxel patterns was performed. The decoder trained on functional maps from other individuals in the common space showed a transfer decoding accuracy comparable to that of an individual decoder trained on single-subject functional maps. The DTI-based registration allowed more precise transformation of gray matter boundaries than a well-established T1-based method. These results suggest that the DTI-based registration is a promising tool for standardization of the brain functions, and moreover, will allow us to perform 'zero-shot' learning of decoders which is profitable in brain machine interface scenes.

Introduction

Functional MRI (fMRI) data is notoriously individualistic; every person’s brain activity patterns are as unique as a fingerprint. This "individuality" poses a massive hurdle for Brain-Machine Interfaces (BMIs) and neuroimaging research. Currently, building a decoder to predict mental states usually requires training on specific individuals. Transferring a decoder typically necessitates a "calibration" phase where the new subject performs the task to fine-tune the model. Standard 3D registration methods rely on T1-weighted anatomical scans focusing on gray matter shape, which doesn't always correlate perfectly with functional connectivity, leading to "blurry" alignments.

Methods

The researchers developed a registration pipeline based on Diffusion Tensor Imaging (DTI). While T1 scans show gray matter "landscapes," DTI maps white matter "highways" by measuring water diffusion direction.

Schematic of the DTI-based registration pipeline.

The pipeline involves:

  1. DTI-based Registration: Using DTI-TK for non-rigid, 3D registration focusing on fiber orientation.
  2. Common Coordinate Space: Mapping 21 subjects into a common space derived from DTI averages.
  3. Decoding Task: Subjects performed a spatial attention task (left vs. right).
  4. Decoder Architectures: A "Conjunction Decoder" (pooled SVM) and a "Naive-Voting Decoder" (ensemble of SVMs).
Training data structure for the multi-subject decoder.

The study used an original database of 22 healthy subjects, collecting T1-weighted structural images, DTI anatomical scans, and fMRI during visual spatial attention tasks.

Key Insights and Results

The core finding is that functional maps are more closely tied to white matter connectivity than to gray matter folds. Aligning the "wiring" leads to a more functionally consistent common space, particularly at the edges of gray matter where neurons connect to long-range fibers.

Key results include:

  • Superior Accuracy: DTI-based registration significantly outperformed T1-based registration (p < 0.01).
  • Zero-Shot Performance: Subject-transfer accuracy (trained on others) was nearly as high as self-decoding accuracy. This allows applying a decoder to a new person using only their structural DTI scan, with no functional calibration required.
  • Robustness: DTI superiority was maintained even with reduced voxel counts.
  • Complementary Information: DTI-based (3D) and surface-based (2D) registrations showed weak correlation, implying they capture different anatomical features.

Discussion

While promising, the study focused on a simple visual task. Future work should explore complex cognitive tasks and population-averaged DTI atlases. A hybrid "multi-modal" registration method combining T1, DTI, and surface data could offer the ultimate solution for brain standardization. Validating these findings on larger datasets like the Human Connectome Project will be crucial for establishing universal decoders.

Summary updated

Abstract

Functional magnetic resonance imaging (fMRI) acquisitions include a great deal of individual variability. This individuality often generates obstacles to the efficient use of databanks from multiple subjects. Although recent studies have suggested that inter-regional connectivity reflects individuality, conventional three-dimensional (3D) registration methods that calibrate inter-subject variability are based on anatomical information about the gray matter shape (e.g., T1-weighted). Here, we present a new registration method focusing more on the white matter structure, which is directly related to the connectivity in the brain, and apply it to subject-transfer brain decoding. Our registration method based on diffusion tensor imaging (DTI) transferred functional maps of each individual to a common anatomical space, where a decoding analysis of multi-voxel patterns was performed. The decoder trained on functional maps from other individuals in the common space showed a transfer decoding accuracy comparable to that of an individual decoder trained on single-subject functional maps. The DTI-based registration allowed more precise transformation of gray matter boundaries than a well-established T1-based method. These results suggest that the DTI-based registration is a promising tool for standardization of the brain functions, and moreover, will allow us to perform 'zero-shot' learning of decoders which is profitable in brain machine interface scenes.
BibTeX
@inproceedings{fuchigami2018zero,
  author = {Takuya Fuchigami and Yumi Shikauchi and Ken Nakae and Manabu Shikauchi and Takeshi OGAWA and Shin Ishii},
  title = {{Zero-shot fMRI decoding with three-dimensional registration based on diffusion tensor imaging}},
  booktitle = {Scientific},
  year = {2018},
  doi = {10.1038/s41598-018-30676-3},
}
Figure