Author: Henry Szu-Meng Chen, Mu-Lan Jen, Vinodh A. Kumar, Ho-Ling Anthony Liu, Jian Ming Teo 👨🔬
Affiliation: School of Medicine, University of Colorado Denver, Department of Neuroradiology, The University of Texas MD Anderson Cancer Center, Department of Imaging Physics, The University of Texas MD Anderson Cancer Center 🌍
Purpose: Resting-state (rs-) fMRI detects functional networks by measuring synchronization of low-frequency oscillations in blood-oxygenation-level-dependent (BOLD) signals between brain regions. Standardization of rs-fMRI processing pipelines is in increasing demand, where a digital reference object (DRO) can serve as a surrogate to validate their software implementation. This study aimed to construct a realistic DRO that can be easily deployed to test clinical rs-fMRI software.
Methods: A brain-like DRO was constructed with structural models and functional atlases (language and hand motor). The BOLD signal was simulated with scanning parameters of a 3T clinical rs-fMRI protocol. Resting-state networks (RSNs) were simulated with spontaneous neuronal responses (pseudo-random events at 0.05 Hz) convolving with hemodynamic response functions. The physiological noise was assigned to simulate respiratory and cardiac responses. Racian noise was added to simulated rs-fMRI images with GM temporal SNR at 50. The DRO was tested using FSL MELODIC (v3.15; FMRIB, Oxford, UK) following a standard processing pipeline, including bandpass filtering, spatial smoothing, and spatial independent component analysis (ICA).
Results: Rs-fMRI DRO was exported as 4D DICOM dataset with embedded RSNs representing language and motor networks from population-based functional probabilistic maps. From RSNs recommended by ICA, two independent networks were successfully identified with minimal distortion of time course (language: r=0.9890; motor: r=0.9929). The spatial localization is reasonably aligned with nominal RSNs (dice similarity coefficient of 0.6158 for language and 0.5716 for motor RSNs at FSL default threshold), suggesting ICA is robust to additive physiological and thermal noises.
Conclusion: This study developed a brain-like DRO that simulates rs-fMRI BOLD responses. The DRO is a DICOM dataset that can be imported to fMRI software. The framework can be useful in validating processing steps and understanding confounding factors in clinical rs-fMRI applications. The DRO can be further expanded to investigate scanning parameters and pre-processing steps for site-specific pipelines.