Signal-to-noise ratio improvement in postmortem MRI using deep learning reconstruction (SwiftMR)

Postmortem imaging has become an essential modality for death investigation, with computed tomography (PMCT) being widely adopted as a rapid and objective screening tool. However, the application of postmortem magnetic resonance imaging (PMMRI) remains limited in routine forensic practice, primarily due to the longer acquisition time and the need for high-quality images under time constraints. SwiftMR™ (AIRS Medical, Seoul, Korea) is a deep learning-based reconstruction software designed to enhance MR image quality through denoising and multi-dimensional k-space optimization, allowing improved signal-to-noise ratio (SNR) without extending scan time. While this technology is commercially available and used clinically, its utility in postmortem imaging has not yet been reported. This study evaluated the effect of SwiftMR on SNR and image quality in postmortem head MRI using a low-field (0.3 T) scanner. Thirty-three consecutive cases (20 males, 13 females; age 33-92 years, mean age 75.2) were analyzed. T1-, T2-, and T2*-weighted images were acquired using a fixed protocol with a total examination time under 15 min. The images were processed with SwiftMR (v3.0.11.0) and analyzed using ROI-based SNR measurements and visual scoring by three blinded observers. SwiftMR significantly increased SNR in all sequences (T1WI, T2WI, and T2*WI; p < 0.001), while mean signal intensity showed no significant difference except for a slight decrease in T2WI, attributed to correction of Rician/Rayleigh bias. Visual scores for image quality were markedly higher after SwiftMR (7.5 ± 1.0 vs. 3.2 ± 0.8; p < 0.001). In conclusion, deep learning-based reconstruction using SwiftMR effectively enhanced image quality and SNR in low-field postmortem MRI without increasing scan time. This approach may facilitate the broader application of postmortem MRI as a practical adjunct to PMCT in routine forensic investigations.