Deep learning reconstruction for liver DWI: impact on image quality and ADC quantification

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Diffusion-weighted imaging (DWI) is crucial for liver lesion assessment.
• Compressed sensing (CS) accelerates DWI acquisition but can impact image quality.
• Model-based deep learning (DL) offers potential for reconstructing high-quality DWI with accelerated acquisition.

Purpose of the Study:

• To compare the image quality, apparent diffusion coefficient (ADC) values, and lesion detection capabilities of model-based deep learning reconstruction for DWI (DL-DWI) against standard compressed sensing DWI (CS-DWI).

Main Methods:

• Retrospective analysis of 188 liver DWI scans (October 2024 - January 2025).
• DL-DWI utilized Adaptive-CS-Net, a model-based deep learning architecture within the CS pipeline.
• Quantitative and qualitative assessments by two radiologists, including lesion conspicuity evaluation.

Main Results:

• DL-DWI demonstrated significantly higher signal-to-noise ratio and contrast-to-noise ratio for lesions compared to CS-DWI.
• Lower coefficient of variation and ADC values were observed in DL-DWI.
• DL-DWI achieved superior image quality scores and significantly enhanced overall lesion conspicuity, particularly for small (<10 mm) and superficial lesions.

Conclusions:

• Model-based DL reconstruction significantly improves both qualitative and quantitative image quality in liver DWI.
• DL-DWI enhances lesion conspicuity, offering a notable advantage for detecting small and superficial liver abnormalities.
• This technique holds promise for more accurate and efficient liver lesion characterization.