Deep Learning Breast DWI Preferred in Retrospective Imaging Study

Single-shot echo-planar imaging with deep learning reconstruction received higher image-quality ratings and stronger reader preference than multi-shot echo-planar imaging with simultaneous multislice in a retrospective breast MRI study. The comparison included 77 breast cancer cases from 74 consecutive women whose mean age was 53.1 years. Researchers also found no statistically significant difference in cancer detection rate between the two acquisition methods, while the single-shot sequence with reconstruction shortened image acquisition.

The study compared ss-EPI with DLR against ms-EPI with SMS for breast diffusion-weighted imaging. Two radiologists assessed qualitative features, including fat suppression, artifacts, lesion conspicuity and overall image quality, as well as their preferred sequence. The investigators also calculated the cancer detection rate. Quantitative analyses examined apparent diffusion coefficient, signal-to-noise ratio, contrast-to-noise ratio and lesion contrast, while a phantom provided an additional comparison of ADC and SNR values. Because the research was retrospective, the results describe performance in the selected clinical cohort rather than establishing superiority through prospective testing.

Qualitative analysis favored ss-EPI with DLR in two important respects: radiologists gave it significantly higher scores for fat suppression and overall image quality. Artifact presence and lesion conspicuity were comparable between the methods. Both radiologists also selected the single-shot sequence with deep learning reconstruction more often than the multi-shot sequence with simultaneous multislice. Reader 1 preferred ss-EPI with DLR in 78.4% of cases, and Reader 2 did so in 89.2%. Despite these clear preferences, the study detected no significant difference in cancer detection rate.

Quantitative analysis found higher lesion contrast with ss-EPI with DLR for both readers, with differences reported at p ≤ 0.001. ADC and SNR values in the clinical images were comparable between the two sequences. Phantom measurements likewise showed no significant differences in ADC and SNR, although the reported SNR values were numerically higher for ss-EPI with DLR than for ms-EPI with SMS: 630.03 ± 759.26 versus 607.45 ± 565.21, respectively, with p = 0.912. The authors therefore distinguished the higher lesion-contrast measurements and stronger qualitative ratings from the broadly similar ADC and SNR results.

Acquisition time also favored the single-shot method. Imaging with ss-EPI and DLR took 2.06 minutes, compared with 3.29 minutes for ms-EPI with SMS, representing an approximately 40% reduction in acquisition time according to the study’s discussion. Taken together, the findings show that ss-EPI with DLR was associated in this retrospective cohort with better fat suppression, higher overall image-quality ratings, greater reader preference, higher measured lesion contrast and a shorter acquisition than ms-EPI with SMS. The methods produced comparable artifact ratings, lesion conspicuity, ADC and SNR measurements, while their cancer detection rates did not differ significantly.

Subject of Research: Retrospective comparison of deep learning reconstruction and simultaneous multislice imaging for breast diffusion-weighted MRI

Article Title: Comparison of single-shot echo-planar imaging with deep learning reconstruction (ss-EPI with DLR) and multi-shot echo-planar imaging with simultaneous multislice (ms-EPI with SMS) for breast DWI: enhanced image quality and reader preference

Article References: Ahn, H. S., Kim, S. H., Hong, M. J., Lee, H.-S., & Benkert, T. (2026). Comparison of single-shot echo-planar imaging with deep learning reconstruction (ss-EPI with DLR) and multi-shot echo-planar imaging with simultaneous multislice (ms-EPI with SMS) for breast DWI: enhanced image quality and reader preference. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02946-4

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02946-4

Keywords: breast MRI, diffusion-weighted imaging, deep learning reconstruction, single-shot EPI, multi-shot EPI, simultaneous multislice, image quality, reader preference, lesion contrast, cancer detection rate, Comparison, single-shot