Enhancing diffusion-weighted prostate MRI through self-supervised denoising and evaluation.

Area of Science:

• Medical Imaging
• Biophysics
• Machine Learning

Background:

• Diffusion-weighted imaging (DWI) is vital for stroke and oncology but suffers from low signal-to-noise ratio (SNR).
• Supervised deep learning denoising requires noise-free training data, which is unavailable for DWI.
• Existing self-supervised methods and conventional denoising techniques have limitations in DWI applications.

Purpose of the Study:

• To develop a self-supervised denoising method for DWI that eliminates the need for ground-truth data.
• To evaluate the denoising performance using a novel self-supervised methodology.
• To demonstrate the potential for accelerating DWI acquisition through reduced repetitions.

Main Methods:

• A self-supervised deep learning approach utilizing an adapted Stein's unbiased risk estimator (SURE).
• Phase-corrected combination of repeated DWI acquisitions to enhance signal and enable self-supervision.
• A self-supervised evaluation metric based on residual signal analysis.

Main Results:

• The proposed method significantly outperforms state-of-the-art self-supervised denoising techniques and conventional non-learning methods.
• The approach successfully denoises DWI scans without requiring noise-free training data.
• Demonstrated feasibility of accelerating DWI scans by acquiring fewer repetitions.

Conclusions:

• Self-supervised learning offers a viable solution for denoising DWI, overcoming the limitations of supervised methods.
• The SURE-based approach provides robust denoising and a reliable evaluation framework for DWI.
• This technique has the potential to improve diagnostic accuracy and efficiency in clinical MRI.