Deep learning based multi-shot breast diffusion MRI: Improving imaging quality and reduced distortion.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Single-shot diffusion-weighted imaging (SS-DWI) is a standard technique for breast MRI.
• Multiplexed sensitivity encoding (MUSE) aims to improve image quality by acquiring data in multiple shots.
• Deep learning (DL) reconstruction offers potential for further enhancement of MRI techniques.

Purpose of the Study:

• To compare the imaging performance of deep-learning reconstructed MUSE (MUSE DL) with SS-DWI for breast MRI.
• To evaluate image quality metrics including signal-to-noise ratio (SNR) and image distortion.
• To assess the impact on apparent diffusion coefficient (ADC) values and diagnostic accuracy.

Main Methods:

• Prospective study involving 61 female participants with 65 breast lesions.
• Acquisition of both SS-DWI and multi-shot MUSE DWI data on a 3T MRI scanner.
• Quantitative analysis of SNR, ADC values, and Hausdorff distance (HD) for distortion assessment.
• Subjective qualitative analysis using Likert scale.

Main Results:

• MUSE DL significantly improved SNR in fibroglandular tissue compared to non-DL MUSE (2-shot DL: 207.8%, 4-shot DL: 175.1%).
• Significantly reduced image distortion was observed with MUSE DL (2-shot: 3.11 mm, 4-shot: 2.58 mm) compared to SS-DWI (4.15 mm).
• No significant difference in ADC values was found between MUSE, MUSE DL, and SS-DWI for benign or malignant tumors.

Conclusions:

• MUSE DL enhances image quality in breast MRI by improving SNR and minimizing distortion.
• The technique preserves the diagnostic accuracy of lesion characterization and ADC values.
• MUSE DL represents a promising advancement for breast DWI.