Accelerated Diffusion-Weighted MRI of Rectal Cancer Using a Residual Convolutional Network.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Diffusion-weighted MRI (DW-MRI) is crucial for rectal cancer diagnosis.
• Acquiring high b-value DW-MRI is time-consuming, limiting clinical application.
• Accelerated acquisition techniques often compromise image quality.

Purpose of the Study:

• To develop and evaluate a deep learning-based denoising technique for accelerating high b-value DW-MRI in rectal cancer.
• To assess the feasibility of using fewer repetitions (NEX) for faster image acquisition.
• To compare the image quality of denoised accelerated scans against the clinical standard.

Main Methods:

• A denoising convolutional neural network (DCNN) with a combined L1-L2 loss function was developed.
• DCNN was trained on 85 rectal cancer patient datasets and tested on 20 datasets.
• Scans were acquired with varying NEX (1, 2, 4), corresponding to acceleration factors of 16, 8, and 4.
• Image quality was qualitatively assessed by expert radiologists.

Main Results:

• Denoised images with 8-fold acceleration (NEX=2) achieved image quality comparable to the clinical standard.
• Denoised images with 4-fold acceleration (NEX=4) surpassed the quality of the clinical standard.
• Expert readers reported similar or improved image quality for accelerated, denoised scans compared to reference scans.
• The technique effectively reduced noise in faster-acquired images.

Conclusions:

• Deep learning-based denoising can significantly accelerate high b-value DW-MRI acquisition for rectal cancer.
• Eightfold acceleration yields similar image quality, while fourfold acceleration provides superior quality compared to standard methods.
• This technique holds promise for improving diagnostic efficiency and accuracy in rectal cancer imaging.