Clinical evaluation of accelerated diffusion-weighted imaging of rectal cancer using a denoising neural network

Background: To evaluate the effectiveness of a deep learning denoising approach to accelerate diffusion-weighted imaging (DWI) and thus improve diagnostic accuracy and image quality in restaging rectal MRI following total neoadjuvant therapy (TNT). Methods: This retrospective single-center study included patients with locally advanced rectal cancer who underwent restaging rectal MRI between December 30, 2021, and June 1, 2022, following TNT. A convolutional neural network trained with DWI data was employed to denoise accelerated DWI acquisitions (i.e., acquisitions performed with a reduced number of repetitions compared to standard acquisitions). Image characteristics and residual disease were independently assessed by two radiologists across original and denoised images. Statistical analyses included the Wilcoxon signed-rank test to compare image quality scores across denoised and original images, weighted kappa statistics for inter-reader agreement assessment, and the calculation of measures diagnostic accuracy. Results: In 46 patients (median age, 60 years [IQR: 47-72]; 37 men and 9 women), 8- and 16-fold accelerated images maintained or exhibited enhanced lesion visibility and image quality compared with original images that were performed 16 repetitions. Denoised images maintained diagnostic accuracy, with conditional specificities up to 96 %. Moderate-to-high inter-reader agreement indicated reliable image and diagnostic assessment. The overall test yield for denoised DWI reconstructions ranged from 76-98 %, demonstrating a reduction in equivocal interpretations. Conclusion: Applying a denoising network to accelerate rectal DWI acquisitions can reduce scan times and enhance image quality while maintaining diagnostic accuracy, presenting a potential pathway for more efficient rectal cancer management.