Impact of a deep-learning image reconstruction algorithm on ventilation and perfusion parameters derived by phase-resolved functional lung (PREFUL) magnetic resonance imaging (MRI) in a prospective healthy cohort

AIM: Deep learning (DL)-based reconstruction algorithms are increasingly integrated into clinical practice, offering the potential for shorter imaging times and improved image quality. This study is the first to evaluate whether DL-reconstructed magnetic resonance imaging (MRI) images affect ventilation and perfusion parameters derived from phase-resolved functional lung (PREFUL) MRI. MATERIALS AND METHODS: This prospective study included 24 healthy volunteers (12 male, mean age: 28.9 ± 12.8 years) with normal lung function confirmed by spirometry. Free-breathing lung MRI was performed at 1.5T using 2D fast spoiled gradient echo sequences. Raw imaging data were reconstructed using both conventional non-DL- and DL-based methods. Image quality was assessed via signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR). PREFUL MRI was used to derive regional ventilation (Rvent), flow-volume loop correlation metric (FVL-CM), perfusion (Q), ventilation and perfusion defect percentages (VDPs and QDPs), and ventilation-perfusion (V/Q) match. Parameters were compared using paired statistical tests and Bland-Altman analysis. RESULTS: DL-based reconstruction significantly improved image quality (SNR: 41.7 ± 19.5 vs 18.0 ± 6.5; contrast-to-noise: 179.9 ± 84.4 vs 75.3 ± 25.7; both P