hide
Free keywords:
-
Abstract:
Time series analysis of heterogeneous preclinical functional magnetic resonance imaging (fMRI) studies poses challenges due to data volume and method heterogeneity. Recent advances in machine learning (ML) and artificial intelligence (AI) allow for addressing such challenges in complex datasets. These approaches can uncover patterns, including temporal kinetics, within blood-oxygen-level-dependent (BOLD) time series with a reduced workload. However, the typically low temporal resolution and signal-to-noise ratio (SNR) of fMRI time series have so far limited progress in this area.
Therefore, we used ultrafast 1D line-scanning and 2D-fMRI data for this study to assess whether enhanced spatial and temporal resolution of fMRI data, along with a sophisticated metric design for clustering, was sufficient to detect differences in BOLD response characteristics in the time domain across cortical layers. Next, we compared consistency of the produced results across four independent studies conducted at two different imaging centers, each utilizing distinct study protocols and, finally, we combined line-scanning data from different studies for time-domain clustering analysis to facilitate a cross-study examination of somatosensory information processing during sensory stimulation of the forepaw.
By adopting a voxel-based and purely data-driven approach, we systematically explored different clustering techniques and analyzed somatosensory cortex fMRI data obtained during forepaw stimulation in rats. We established and validated an unsupervised workflow capable of detecting BOLD response latencies between different stimulus modalities, producing consistent results across different study protocols, indicating robustness, reproducibility, and generalizability of our framework.
