Deep Learning-Enhanced T1-Weighted Imaging for Breast MRI at 1.5T

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Breast MRI is crucial for cancer diagnosis and management.
• Standard T1w VIBE sequences can be limited by image quality and artifacts.
• Advancements in deep learning offer potential to enhance MRI sequences.

Purpose of the Study:

• To assess the image quality of a novel deep-learning-based T1w volumetric interpolated breath-hold (VIBE_DL) sequence.
• To compare VIBE_DL with the standard T1w VIBE (VIBE_Std) sequence in breast MRI.
• To evaluate diagnostic confidence and lesion characterization using both sequences.

Main Methods:

• Prospective study involving 52 breast cancer patients at 1.5T MRI.
• Acquisition of both T1w VIBE_Std and T1 VIBE_DL sequences (pre- and post-contrast).
• Independent assessment by two radiologists using a Likert scale for image quality, sharpness, artifacts, SNR, and diagnostic confidence.

Main Results:

• VIBE_DL demonstrated significantly higher scores for image quality, sharpness, SNR, and diagnostic confidence compared to VIBE_Std (p < 0.001).
• Post-contrast VIBE_DL and its subtraction (SUB_DL) also showed superior image quality, sharpness, SNR, and reduced artifacts (p < 0.001).
• SUB_DL provided superior diagnostic certainty for one reader (p = 0.004).

Conclusions:

• Deep learning-enhanced T1w VIBE_DL significantly improves image quality in 1.5T breast MRI.
• VIBE_DL offers enhanced sharpness, SNR, and diagnostic confidence over standard VIBE_Std.
• This DL-based sequence represents a valuable advancement for breast MRI diagnostics.