Oral Abstract 1: Initial Evaluation of Deep Resolve Boost and Sharp for in vivo Cardiac Diffusion Tensor Imaging

Diffusion tensor imaging (DTI) provides a non-invasive tool to assess cardiac tissue microstructure^1-5. However, low signal-to-noise ratio (SNR) and long acquisition times remain challenges. AI-based Deep Resolve technology (Siemens Healthineers, Forchheim, Germany) enables denoising and improving resolution using raw-data-to-image deep neural network. While it has been successfully used for neuroimaging, MSK, and other applications⁶, its effect on cardiac DTI remains unknown. This study aims to apply Deep Resolve Boost (DRB) and Deep Resolve Sharp (DRS) applications to cardiac DTI and evaluate its effect on SNR, image quality, and diffusion parameter maps.

Methods:

Five healthy volunteers and 4 patients were scanned under an IRB approved protocol on a 3T MRI scanner (MAGNETOM Cima.X, Siemens Healthineers, Forchheim, Germany). Images were acquired at end systole with ECG triggering under free-beathing using an M2 compensated⁷ SE-EPI research sequence with inner volume excitation, FOV 350mm×131mm, matrix size 128×48, 8mm slice, 5 slices, 12 diffusion directions, averages: 1 for b=50s/mm² and 8 for b=500s/mm². Total acquisition time was < 10min depending on subjects’ heart rate. DRB (denoising level medium) and DRS were enabled prospectively during the acquisition. The same raw data was retrospectively reconstructed on the scanner with DRB denoising level high plus DRS, as well as DRB and DRS off for comparison.

A custom-built software in MATLAB was used to perform motion correction⁸ and generate mean diffusivity (MD), fractional anisotropy (FA), color FA, and helix angle (HA) maps. Mean and standard deviations of MD and FA values were calculated by manually segmenting the left ventricle (LV). SNR was calculated as mean signal intensity in LV divided by standard deviations of signal intensity in a background ROI averaged across all diffusion weighted images (DWI) for each b value. Diffusion maps generated using averages of 8, 4, 2, and 1 for b=500 s/mm² DWI were compared in one volunteer.

Results:

DRB achieved significant SNR enhancement for both b-value images especially with denoising level high (Fig1). Fig2 shows DWI and diffusion parameter maps of one slice in one volunteer. DWI of both b-values are visually sharper with DRS as observed around myocardium borders and papillary muscles (Fig2A). DRB and DRS preserved smoothly varying helical structure even with fewer averages for high b-value acquisition; while disruption was observed without DRB and DRS (indicated by arrows in Fig2B). Table 1 shows comparison of global MD and FA values in all subjects. A trend of slight increase in MD and FA values was observed using DRB and DRS.

Conclusion:

In this preliminary study, enhancement in SNR and image sharpness was observed in human cardiac DTI images using DRB and DRS. Global FA and MD values obtained with DRB and DRS are comparable to those without it with a slight increasing trend. Further investigations will be performed in larger healthy and patient cohorts for statistical analysis.