Independent Component Analysis for Non-Contrast-Enhanced Free-Breathing Pulmonary Ventilation and Perfusion MRI

To introduce a robust approach for non-contrast-enhanced free-breathing functional lung MRI based on independent component analysis (ICA), addressing key limitations of standard methods such as sensitivity to irregular breathing or irregular heartbeat.

Repeated dynamic single-slice MRI of ten healthy volunteers and twenty-five lung tumor patients was performed in free breathing on a $0.35T$ MR-linac using a 2D coronal cine bSSFP sequence, resulting in 143 datasets. Following motion correction, lung segmentation, and signal alignment, temporal ICA was applied to extract ventilation and perfusion components. Ventilation- and perfusion-weighted maps were generated and compared to Fourier decomposition (FD) and matrix pencil (MP) analysis. Signal construction performance was evaluated using the Pearson correlation coefficient and nonparametric statistical testing, comparing the methods under regular and irregular breathing and heartbeat conditions. Map accuracy and robustness at low signal-to-noise ratio (SNR) were additionally assessed in numerical-phantom simulations.

In vivo, ICA effectively separated physiological lung signals, achieving significantly higher temporal agreement with original voxel time courses than both FD and MP (all $p<0.001$ ). Under all physiological conditions, ICA yielded the highest temporal correlations. In simulations, ICA maintained robust signal fidelity, especially under irregular breathing and low-SNR conditions, preserving $>90\%$ of contrast in simulations, while FD and MP exhibited pronounced degradation and significantly reduced temporal correlations.

ICA functional lung MRI provides a robust alternative to conventional techniques, enabling reliable ventilation and perfusion assessment from free-breathing acquisitions, particularly in the presence of irregular respiratory patterns and motion.