Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

JUMP: A joint multimodal registration pipeline for neuroimaging with minimal preprocessing

Adria Casamitjana, Juan Eugenio Iglesias, Raul Tudela, Aida Ninerola-Baizan, Roser Sala-Llonch

TL;DR

This work tackles the fragmentation and potential biases of multimodal neuroimaging pipelines by introducing JUMP, a joint multimodal registration pipeline with minimal preprocessing. It employs an orbit-like graph structure linking a session-specific template to multiple modalities through latent transforms and a set of pairwise rigid registrations, parameterized in the Lie algebra with $\mathcal{R}_k=\exp[\bm{R}_k]$ and $\mathcal{T}_n=\exp[\bm{T}_n]$ under a Laplace likelihood. The approach enables fast, scalable inference on large cohorts while producing modality-specific biomarkers from T1w, rs-fMRI, and amyloid PET data, validated on ADNI with DMN analyses, SUVr comparisons, and cross-modal correlations. Code is publicly available, positioning the method as a scalable, unbiased platform for multimodal neuroimaging studies in dementia.

Abstract

We present a pipeline for unbiased and robust multimodal registration of neuroimaging modalities with minimal pre-processing. While typical multimodal studies need to use multiple independent processing pipelines, with diverse options and hyperparameters, we propose a single and structured framework to jointly process different image modalities. The use of state-of-the-art learning-based techniques enables fast inferences, which makes the presented method suitable for large-scale and/or multi-cohort datasets with a diverse number of modalities per session. The pipeline currently works with structural MRI, resting state fMRI and amyloid PET images. We show the predictive power of the derived biomarkers using in a case-control study and study the cross-modal relationship between different image modalities. The code can be found in https: //github.com/acasamitjana/JUMP.

JUMP: A joint multimodal registration pipeline for neuroimaging with minimal preprocessing

TL;DR

This work tackles the fragmentation and potential biases of multimodal neuroimaging pipelines by introducing JUMP, a joint multimodal registration pipeline with minimal preprocessing. It employs an orbit-like graph structure linking a session-specific template to multiple modalities through latent transforms and a set of pairwise rigid registrations, parameterized in the Lie algebra with and under a Laplace likelihood. The approach enables fast, scalable inference on large cohorts while producing modality-specific biomarkers from T1w, rs-fMRI, and amyloid PET data, validated on ADNI with DMN analyses, SUVr comparisons, and cross-modal correlations. Code is publicly available, positioning the method as a scalable, unbiased platform for multimodal neuroimaging studies in dementia.

Abstract

We present a pipeline for unbiased and robust multimodal registration of neuroimaging modalities with minimal pre-processing. While typical multimodal studies need to use multiple independent processing pipelines, with diverse options and hyperparameters, we propose a single and structured framework to jointly process different image modalities. The use of state-of-the-art learning-based techniques enables fast inferences, which makes the presented method suitable for large-scale and/or multi-cohort datasets with a diverse number of modalities per session. The pipeline currently works with structural MRI, resting state fMRI and amyloid PET images. We show the predictive power of the derived biomarkers using in a case-control study and study the cross-modal relationship between different image modalities. The code can be found in https: //github.com/acasamitjana/JUMP.
Paper Structure (14 sections, 6 equations, 4 figures, 1 table)

This paper contains 14 sections, 6 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Material and methods
  3. Registration framework
  4. Pipeline
  5. Initial segmentation
  6. Rigid transform
  7. Minimal preprocessing for downstream tasks
  8. Dataset
  9. Experiments and results
  10. Resting-state fMRI
  11. Amyloid PET
  12. Multimodal analysis
  13. Discussion and future work
  14. Acknowledgments

Figures (4)

  • Figure 1: The "orbit-like" graph structure chosen. In the center, the session-specific template and the $\{m_i\}_{i=1}^N$ modalities around. In red, the dense observational graph built from pairwise registration of all modalities, while in black our choice of spanning tree.
  • Figure 2: Top: we show the default mode network (DMN) in MNI space computed using Group-ICA over all subjects' time-series. Bottom: we compare baseline DMN amplitude for $N_{\text{CN}}=162$, $N_{\text{AD}}=30$.
  • Figure 3: Relationship between PET standard uptake value for the parietal cortex and CSF biomarkers for the entire population (left column; $N=396$) and stable CN and AD subjects (right column; $N_{\text{CN}}=68$, $N_{\text{AD}}=45$). We overlay a local regression function to show the tendency of the relationship.
  • Figure 4: Pairwise relationship between different biomarkers available in the dataset ($N=152$). Concretely, we use CSF p-tau and image biomarkers derived from different modalities (T1w MRI, rs-fMRI and Amyloid PET). In the diagonal, we plot the distribution of each biomarker.