Reduction of ADC bias in diffusion MRI with deep learning-based acceleration: A phantom validation study at 3.0 T

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence in Medicine

Background:

• Accelerated diffusion-weighted imaging (DWI) is desirable in diagnostic radiology but faces challenges with low signal-to-noise ratio (SNR) in high b-value images.
• Low SNR in DWI leads to bias and variability in quantitative apparent diffusion coefficient (ADC) values, impacting diagnostic accuracy.
• Deep learning-based reconstruction and denoising offer potential solutions to overcome SNR limitations in accelerated DWI.

Purpose of the Study:

• To investigate the impact of SNR reduction on ADC bias and variability in DWI.
• To evaluate the performance of deep learning-based reconstruction (C-SENSE AI) compared to conventional (SENSE) and compressed sensing (C-SENSE) methods.
• To assess the effectiveness of C-SENSE AI in reducing ADC bias and random measurement error under accelerated conditions.

Main Methods:

• Investigated SNR reduction effects on ADC bias and variability using a diffusion phantom and numerical simulations.
• Compared reconstruction methods: SENSE, C-SENSE, and C-SENSE AI at varying acceleration factors and flip angles.
• Assessed ADC bias using Lin's Concordance Correlation Coefficient (CCC) and random measurement error (RME) using the mean coefficient of variation (CV¯).

Main Results:

• Simulations and phantom measurements confirmed that lower SNR (due to increased acceleration or decreased flip angle) leads to increased ADC bias and RME.
• C-SENSE AI reconstruction demonstrated superior performance, yielding significantly improved ADC maps with reduced bias and error compared to SENSE and C-SENSE.
• At high acceleration and low flip angle, C-SENSE AI achieved CCC values of 0.987 and CV¯ of 0.254, significantly outperforming other methods.

Conclusions:

• Deep learning-based C-SENSE AI reconstruction effectively mitigates ADC bias and random measurement error in accelerated DWI.
• C-SENSE AI offers a promising solution for improving quantitative accuracy in low SNR DWI, enabling further acceleration for diagnostic applications.
• The study highlights the potential of AI in enhancing the reliability and efficiency of quantitative diffusion imaging.