Imaging analysis

The spectral-based image data of the three phases were postprocessed with a dedicated workstation (IntelliSpace Portal; Philips Electronics Japan). Two radiologists (D.T. and N.T.) identified the lesions on conventional 120 keV CT images in consensus using endoscopy records as references. They also in consensus evaluated qualitative CT and quantitative dlDECT parameters. The quantitative CT parameters included contour (round or lobulated), growth pattern (endoluminal or exophytic), enhancement pattern (homogeneous, heterogeneous), and presence or absence of necrosis. As a quantitative dlDECT analysis, the readers placed the largest-possible circular region of interest (ROI) that avoided vessels, calcifications, or cystic lesions. The distribution of ROI included three in cardia, four in fundus, and 19 in body. Another ROI was placed in the aorta of the same slice in each case for preparing normalization. The mean iodine density (mg/ml) of the lesion and aorta was calculated, and physiological variations in patients could be minimized using the normalized iodine concentration (NIC) = IC_lesion/IC_aorta. The spectral slope (λ_HU) was adopted as a second dlDECT parameter. The spectral slope reflects the variation of the material CT value with the energy of the X-ray and absorption characteristics relative to different X-ray energies [12]. We calculated the slope using the values of 40 keV and 70 keV monochromatic images according to the following formula: λ_HU = (CT40 keV–CT70 keV)/30. These measurement processes were performed for all three phases.