Deep Learning Denoising Improves CT Perfusion Image Quality in the Setting of Lower Contrast Dosing: A Feasibility Study

• DOI: 10.3174/ajnr.a8367
• Publication Date: 2024-06-06T19:35:14Z
• Authors: Mahmud Mossa-Basha; Chengcheng Zhu; Tanya Pandhi; Steve Mendoza; Javid Azadbakht; Ahmed Safwat; Dean Homen; Carlos Zamora; Dinesh Kumar Gnan...; Ruiyue Peng; Steven Cen; Vinay Duddalwar; Jeffry R. Alger; Danny J.J. Wang

<h3>ABSTRACT</h3> <h3>BACKGROUND AND PURPOSE:</h3> Considering recent iodinated contrast media (ICM) shortages, this study compared reduced ICM and standard dose CTP acquisitions, the impact of deep learning (DL)-denoising on image quality in preclinical clinical studies. <h3>MATERIALS METHODS:</h3> Twelve swine underwent 9 exams each, performed at combinations 3 different X-ray (37, 67, 127mAs) doses (10, 15, 20mL). Clinical acquisitions before during shortage protocol change (from 40 mL to 30 mL) were retrospectively included. Eleven patients with 11 propensity-score-matched controls A Residual Encoder-Decoder Convolutional-Neural-Network (RED-CNN) was trained for denoising using K-space-Weighted Image Average (KWIA) filtered images as target. The standard, RED-CNN denoised, KWIA noise-filtered animal human studies quantitative SNR qualitative evaluation. <h3>RESULTS:</h3> decreased reductions mAs doses. Contrast reduction had a greater effect than reduction. Noise-filtering by progressively improved maps, resulting highest SNR. generally lower dose, which (p&lt;0.05). Qualitative readings consistently rated denoised best quality, followed then images. <h3>CONCLUSIONS:</h3> DL-denoising can improve low protocols, could approximate CTP, addition potentially improving acquisitions. ABBREVIATIONS: ICM=iodinated media; DL=deep learning; KWIA=k-space weighted average; LCD=low-contrast dose; SCD=standard RED-CNN=Residual Convolutional Neural Network; PSNR=Peak Signal Noise Ratio; RMSE=Root Mean Squared Error; SSIM=Structural Similarity Index.