Comparing DWI image quality of deep-learning-reconstructed EPI with RESOLVE in breast lesions at 3.0T: a pilot study

TL;DR

This pilot study compares deep-learning reconstructed DWI with RESOLVE in breast imaging at 3.0T, showing potential for improved image quality and resolution without longer scan times.

Contribution

It introduces a deep learning-based DWI reconstruction method and evaluates its image quality against the standard RESOLVE technique in breast MRI.

Findings

DWI DL showed significantly different SNR and CNR compared to RESOLVE.

DWI DL may improve spatial resolution without increasing scan time.

Potential for enhanced breast imaging using deep learning reconstruction.

Abstract

The challenging spatial resolution of DWI could be addressed by deep learning based image reconstruction, by reducing noise without increasing acquisition time. To compare the image quality of the Echo Planar Imaging Deep Learning (EPI DL) DWI sequence with the clinically used simultaneous multi slice (SMS) RESOLVE in breast lesions. EPI DL and RESOLVE breast images from 20 participants were qualitatively evaluated. Quantitative image quality metrics of SNR and CNR on both high b-value (b800) images and ADC maps were calculated. SNR in RESOLVE vs. EP DL differed statistically significantly in manually delineations for b800 (p=0.006), ADC maps (p=0.001), and in ADC circularly delineations (0.001). DWI DL reconstruction may be clinically useful for addressing low-spatial resolution without compromising acquisition time and image quality. Such benefits coupled with the available methods of readout segmentation and SMS acquisitions may further enhance the value of DWI in breast imaging.