RT Journal Article
SR Electronic
T1 Deep Learning Denoising Improves CT Perfusion Image Quality in the Setting of Lower Contrast Dosing: A Feasibility Study
JF American Journal of Neuroradiology
JO Am. J. Neuroradiol.
FD American Society of Neuroradiology
SP 1468
OP 1474
DO 10.3174/ajnr.A8367
VO 45
IS 10
A1 Mossa-Basha, Mahmud
A1 Zhu, Chengcheng
A1 Pandhi, Tanya
A1 Mendoza, Steve
A1 Azadbakht, Javid
A1 Safwat, Ahmed
A1 Homen, Dean
A1 Zamora, Carlos
A1 Gnanasekaran, Dinesh Kumar
A1 Peng, Ruiyue
A1 Cen, Steven
A1 Duddalwar, Vinay
A1 Alger, Jeffry R.
A1 Wang, Danny J.J.
YR 2024
UL http://www.ajnr.org/content/45/10/1468.abstract
AB BACKGROUND AND PURPOSE: Considering recent iodinated contrast shortages and a focus on reducing waste, developing protocols with lower contrast dosing while maintaining image quality through artificial intelligence is needed. This study compared reduced iodinated contrast media and standard dose CTP acquisitions, and the impact of deep learning denoising on CTP image quality in preclinical and clinical studies. The effect of reduced X-ray mAs dose was also investigated in preclinical studies.MATERIALS AND METHODS: Twelve swine underwent 9 CTP examinations each, performed at combinations of 3 different x-ray (37, 67, and 127 mAs) and iodinated contrast media doses (10, 15, and 20 mL). Clinical CTP acquisitions performed before and during the iodinated contrast media shortage and protocol change (from 40 to 30 mL) were retrospectively included. Eleven patients with reduced iodinated contrast media dosages and 11 propensity-score-matched controls with the standard iodinated contrast media dosages were included. A residual encoder-decoder convolutional neural network (RED-CNN) was trained for CTP denoising using k-space-weighted image average filtered CTP images as the target. The standard, RED-CNN-denoised, and k-space-weighted image average noise-filtered images for animal and human studies were compared for quantitative SNR and qualitative image evaluation.RESULTS: The SNR of animal CTP images decreased with reductions in iodinated contrast media and milliampere-second doses. Contrast dose reduction had a greater effect on SNR than milliampere-second reduction. Noise-filtering by k-space-weighted image average and RED-CNN denoising progressively improved the SNR of CTP maps, with RED-CNN resulting in the highest SNR. The SNR of clinical CTP images was generally lower with a reduced iodinated contrast media dose, which was improved by the k-space-weighted image average and RED-CNN denoising (P < .05). Qualitative readings consistently rated RED-CNN denoised CTP as the best quality, followed by k-space-weighted image average and then standard CTP images.CONCLUSIONS: Deep learning-denoising can improve image quality for low iodinated contrast media CTP protocols, and could approximate standard iodinated contrast media dose CTP, in addition to potentially improving image quality for low milliampere-second acquisitions.AIartificial intelligenceDLdeep learningICMiodinated contrast mediaKWIAk-space-weighted image averageLCDlow contrast dosePSNRpeak SNRRED-CNNresidual encoder-decoder convolutional neural networkRMSEroot mean square errorSCDstandard contrast doseSSIMstructural similarity index measure