2332
Deep Learning Denoising to Accelerate Diffusion-Weighted Imaging of Rectal Cancer
Mohaddese Mohammadi1, Elena Kayee1, Youngwook Kee1, Jennifer Golia Pernicka 2, Iva Petkovska2, and Ricardo Otazo 2
1Medical physics, Memorial Sloan Kettering Cancer Center, New York, NY, United States, 2Memorial Sloan Kettering Cancer Center, New York, NY, United States
Both the denoised high b-value image and resulting denoised ADC map compare favorably to the original noisy results and approximate the results obtained with the reference image. This result indicates that accelerating rectal DWI by reducing the number of acquired averages is feasible. 
Figure 3. High b-value DWI and apparent diffusion coefficient (ADC) for noisy (NEX=4), denoised DnCNN (NEX=4), and reference (NEX=16). DnCNN denoises the original NEX=4 images and provides results that are closer to the reference (NEX=16). The arrow is pointing to the tumor location.
Figure 2. Application to the DnCNN to a test case. The inputs are the high b-value noisy image to be denoised and the low b-value image used for guidance. The estimated residual noise is subtracted from the noisy image to generate the denoised image.