Deep learning based generation of 3D-maximum Intensity Projection (MIP) of MRA from T1-weighted MRI scans for...

Cerebrovasculature abnormalities are crucial for early vascular disease (VD) detection in the brain, with maximum intensity projection of magnetic resonance angiography (MIP-MRA) being important for its diagnosis. As compared to routine T1-weighted (T1w) MR scans, MIP-MRA has limitations, such as the long acquisition time for MRA and sometimes the need for contrast agents. Hence, this study proposes the synthesis of 3D MIP-MRA images from 3D T1w using the U-Net framework that is trained and tested with the IXI dataset, aiming to leverage advanced synthetic MRA for characterizing VD.The experimental results showed that the model achieved promising performance on the validation set, with a peak signal-to-noise ratio (PSNR) of 32.4516 ± 0.1468 and structural similarity index (SSIM) of 0.8219 ± 0.0016. The framework demonstrated high generalizability in evaluations with Bullitt and Study Forrest datasets, proved scanner-insensitive with SIMON datasets and successfully identified vessel narrowing and discontinuity in ischemic stroke subjects from the ATLAS dataset.The findings open possibilities for routine use of 3D MIP scans in clinical settings, for early VD detection, treatment evaluation, and management, enhancing diagnostic and improving patient outcomes with timely interventions.