A Compact Implicit Neural Representation for Efficient Storage of Massive 4D Functional Magnetic Resonance Imaging

TL;DR

The paper addresses the challenge of compressing massive 4D fMRI data by introducing an implicit neural representation (INR) framework tailored to fMRI's structure. It decomposes neural activation into a set of reusable patterns, models their spatial distributions with INR blocks, and initializes patterns with ICA to preserve salient activity. The approach combines intra-region correlation modeling, activation-pattern decomposition, and nonlinear fusion, achieving superior fidelity (e.g., via PSNR/SSIM) and better preservation of downstream fMRI tasks compared with state-of-the-art methods. This work enables high-fidelity, low-bandwidth sharing of large fMRI datasets and provides a foundation for future INR-based biomedical data compression.

Abstract

Functional Magnetic Resonance Imaging (fMRI) data is a widely used kind of four-dimensional biomedical data, which requires effective compression. However, fMRI compressing poses unique challenges due to its intricate temporal dynamics, low signal-to-noise ratio, and complicated underlying redundancies. This paper reports a novel compression paradigm specifically tailored for fMRI data based on Implicit Neural Representation (INR). The proposed approach focuses on removing the various redundancies among the time series by employing several methods, including (i) conducting spatial correlation modeling for intra-region dynamics, (ii) decomposing reusable neuronal activation patterns, and (iii) using proper initialization together with nonlinear fusion to describe the inter-region similarity. This scheme appropriately incorporates the unique features of fMRI data, and experimental results on publicly available datasets demonstrate the effectiveness of the proposed method, surpassing state-of-the-art algorithms in both conventional image quality evaluation metrics and fMRI downstream tasks. This work in this paper paves the way for sharing massive fMRI data at low bandwidth and high fidelity.

Figures (8)

• Figure 1: The trace and slice comparison of an exemplar fMRI data modeled by previous INR-based methods (NeRF and HNeRV) and the proposed approach.
• Figure 2: The workflow and the basic structure of the proposed compression approach.
• Figure 3: An example of neuronal activation patterns and their distributions.
• Figure 4: The network structure of the Down Block and Up Block.
• Figure 5: The performance of the proposed approach and the benchmark algorithms.