Reduced-dose CT scan of colorectal cancer
摘要
Assessing the objective and subjective image attributes of post-adaptive statistical iterative reconstruction (ASIR-V60%, AV60) and medium-strength deep learning image reconstruction (DLIRM) virtual monoenergetic images (VMI) across various kiloelectron volts (keV), in comparison to polychromatic images (PEI), among colorectal cancer (CRC) patients.
MethodsA total of 69 patients with CRC and 35 volunteers without intestinal diseases were included in the spectral CT scans. Portal vein phase spectral CT scans can be divided into standard- and reduced- dose (RD)-scanning models. PEI and 11 VMI images were reconstructed at every 10 keV interval of 40-140 keV. Density analysis was performed to calculate the signal to noise ratio (SNR) of the normal colonic wall and CRC tissues. For subjective quality assessment, three radiologists independently and double-blind assessed the images using a five-point Likert scale.
ResultsDensity distribution analysis showed the largest attenuation gradient between CRC and normal tissue at 40 keV-VMI (VMI40keV) in standard-dose AV60. Except for VMI50keV, VMI40keV had the highest SNR [18.73 (14.07, 25.75)] and contrast-to-noise ratio (CNR: 9.51±4.42) compared with VMI60-140keV and PEI [SNR: 10.14 (8.56, 12.75); CNR: 3.46±1.64] (all P < 0.001). Compared with PEI, there were statistical differences in SNR of RD-DLIRM and RD-AV60 [SNR: 8.88±1.98 and 7.79 (6.64, 9.70), respectively] lesions (P = 0.017 and 0.005). When VMI40keV was added to PEI, the overall area under the curve of the three radiologists significantly improved from 0.885 to 0.941 for CRC diagnosis (P < 0.001).
ConclusionCompared with PEI, the standard-dose AV60-VMI40keV had better objective image quality and lesion detection. Meanwhile, the image quality of RD-VMI40keV can be guaranteed by DLIRM and can be used for CT screening of CRC.