US 12,333,686 B1
Zero-shot low-dose CT image denoising method and apparatus based on strip diffusion model
Xiangyun Hu, Hubei (CN); Bo Su, Hubei (CN); Yunfei Zha, Hubei (CN); Jun Wan, Hubei (CN); and Jiabo Xu, Hubei (CN)
Assigned to HUBEI LUOJIA LABORATORY, Hubei (CN)
Filed by HUBEI LUOJIA LABORATORY, Hubei (CN)
Filed on Jan. 13, 2025, as Appl. No. 19/017,812.
Claims priority of application No. 202411245470.4 (CN), filed on Sep. 6, 2024.
Int. Cl. G06T 5/60 (2024.01); G06T 5/70 (2024.01)
CPC G06T 5/60 (2024.01) [G06T 5/70 (2024.01); G06T 2207/10081 (2013.01); G06T 2207/20021 (2013.01); G06T 2207/20076 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30004 (2013.01); G06T 2207/30168 (2013.01)] 10 Claims
OG exemplary drawing
 
1. A zero-shot low-dose computed tomography (CT) image denoising method based on strip diffusion model, characterized by comprising:
obtaining a normal-dose CT image under a conventional scanning condition, and based on a computed tomography dose index (CTDI) value of the normal-dose CT image under the conventional scanning condition, obtaining low-dose CT images with CTDI values being a preset ratio of the CTDI value of the normal-dose CT image and obtaining low-dose CT images with different thicknesses;
building a strip diffusion model, the strip diffusion model comprising a forward diffusion structure and a backward inference structure; wherein the forward diffusion structure is configured to add noise gradually to an input normal-dose image by Markov chain to obtain a pure noisy image, and the backward inference structure is configured to divide an input CT image into different strips based on UNet model and at the same time, by backward inference, denoise each strip gradually to generate a denoised strip sequentially, and splice the generated denoised strips into a denoising result; wherein an overlap rate is set between the strips;
inputting the obtained normal-dose CT image into the strip diffusion model, and by mean square loss function, calculating a difference between the noise added by the forward diffusion structure and a noise predicted by the UNet model to train the entire strip diffusion model;
inputting the obtained low-dose CT images with the CTDI values being the preset ratio of the CTDI value of the normal-dose CT image as test data into the trained strip diffusion model to obtain a denoised low-dose CT image.