Shortening 3D T2-Weighted Breast MRI scan time using deep learning based reconstruction: A phantom and patient reader study

Motivation: 3D fast-spin-echo T2w imaging provides higher spatial resolution and wider coverage compared to 2D T2w imaging. However, 3D FSE requires longer scan time due to limited echo trains. Goal(s): To evaluate the use of pseudo-random under-sampled acquisition coupled with deep learning-based reconstruction (Sonic DL) to reduce the scan time while maintaining image quality. Approach: Quantitative analysis was performed on phantom images, and a qualitative reader study was conducted on in-vivo breast images acquired with both standard T2w imaging using Sonic DL. Results: Sonic DL with acceleration factor of 10 reduced scan time by 50% while improving signal-to-noise ratio, sharpness, and overall image quality. Impact: This work demonstrated the feasibility of using a pseudo-random under-sampled acquisition coupled with deep learning-based reconstruction (Sonic DL) to reduce 3D T2w breast MRI scan time by 50% while improving image quality.