第一部分：

目的：探讨对比增强优化技术（contrast enhanced boost, CE-boost）联合混合迭代重建（Hybrid IR，即HIR [AIDR 3D，三维自适应迭代]）和基于模型的迭代重建（MBIR [FIRST，全模型迭代算法]）算法能否改善腹部CT血管造影（CTA）的图像质量。

材料和方法：本回顾性研究纳入2020年5月至8月期间在我院佳能640层CT(Aquilion ONEGENESIS,Canon Medical System Corporation,Japan)接受腹部增强CTA检查的患者50例。动脉期和门静脉期图像分别采用三种不同的算法独立重建[滤波反投影（FBP）、三维自适应迭代（AIDR 3D）和全模型迭代算法（FIRST）]，此外，应用对比增强优化技术（CE-boost）生成AIDR 3D-boost和FIRST-boost图像。比较五组数据集（FBP、AIDR 3D、FIRST、AIDR 3D-boost 和 FIRST-boost）在腹主动脉、肝右动脉、肠系膜上动脉、肠系膜上动脉一级分支和肾动脉、肝右门静脉、门静脉右后分支、肠系膜上静脉和肠系膜上静脉二级分支的信噪比（SNR）及对比噪声比（CNR）的差异，SNR、CNR计算公式如下：信噪比（SNR）=CT值_血管/SD_血管，对比噪声比（CNR）=（ CT值_血管- CT值_肌肉）/SD_脂肪；比较五组数据集动脉期和门静脉期的噪声差异（以前腹壁皮下脂肪CT值的标准差计算）。主观评价方面，两名放射科医生根据总体图像质量、尤其是远端动脉或门脉血管分支的可见性与连续性等情况对5组图像进行独立评分排序（最好：5分，第二：4分，第三：3分，第四：2分，第五：1分）。采用Friedman检验和Dunn-Bonferroni事后检验进行统计分析。

结果：与FBP、AIDR 3D、FIRST及AIDR 3D-boost图像相比，FIRST-boost的动脉和门静脉图像噪声最低（均P < 0.05），且信噪比（SNR）和对比噪声比（CNR）显著高于FBP、AIDR 3D和FIRST图像（均P < 0.05）。AIDR 3D-boost图像的噪声、SNR和CNR均优于FBP及AIDR 3D图像（均P < 0.05）。在主观评分中，FIRST-boost图像评分显著高于FBP、AIDR 3D和AIDR 3D-boost图像（均P < 0.05）。

结论：对比增强优化技术（CE-boost）可改善腹部CTA的图像质量。与其他四组图像数据集相比，基于模型的迭代重建（MBIR）联合CE-boost技术生成的图像（FIRST-boost）具有最佳图像质量。

第二部分：

目的：通过比较多种重建算法，探讨深度学习重建对增强CT上胆道系统图像质量的影响。

材料和方法：本回顾性研究纳入30例2021年3月-2021年4月在我院佳能640层CT(Aquilion ONEGENESIS,Canon Medical System Corporation,Japan)进行增强CT检查并伴有胆总管或肝外胆管扩张的患者。门脉期图像分别采用滤波反投影算法（FBP）、三维自适应迭代（AIDR3D）、全模型迭代算法（FIRST）和深度学习重建（DLR）对门脉期图像进行重建。比较四种重建图像（DLR、FIRST、AIDR3D、FBP）在胆总管、肝实质感兴趣区域的信噪比（SNR）、对比噪声比（CNR）及噪声（以前腹壁皮下脂肪CT值的标准差计算），SNR、CNR计算公式如下：SNR_胆管内=CT值_胆管内/SD_胆管内, CNR=(CT值_胆管内−CT值_肌肉)/SD_脂肪，SNR肝=CT值_肝/SD_肝,CNR=(CT值_肝−CT值_肌肉)/SD_脂肪。在主观图像质量评估中，两名放射科医师根据总体图像质量对4组图像质量进行独立评价排序，将4组图像的图像质量从高到低依次排序，最好：4分，第二：3分，第三：2分，第四：1分。采用Friedman检验和Dunn-Bonferroni事后检验进行统计学比较。

结果：DLR图像的CNR（胆管：4.42±0.87，肝实质：3.78±1.47）显著高于FBP（胆管：2.21±1.02，肝实质：1.43±1.29）、AIDR3D（胆管：2.81±0.91，肝实质：2.39±1.94）及FIRST（胆管：2.51±1.24，肝实质：2.45±1.81）（P均<0.05），DLR图像的SNR（胆管：1.39±0.85，肝实质：9.75±1.90）显著高于FBP（胆管：0.86±0.63，肝实质：3.31±1.12）和FIRST（胆管：1.01±0.61，肝实质：5.73±1.37），DLR图像的噪声（10.51±3.53）显著低于FBP（24.10±3.92）,AIDR3D（15.72±2.41）和FIRST（17.20±3.82）（P均<0.05）。DLR图像的主观评价排序（4分）显著高于FPB（1分）、AIDR3D[2.85（最小值：2；最大值：3]和FIRST[2.15（最小值：2；最大值：3]（P均<0.05）。

结论：深度学习重建能够降低图像噪声、提高增强CT图像质量，有助于对胆道系统进行更好地观察。

第三部分：

目的：评估深度学习重建（DLR）算法对腹部CT图像中壶腹及壶腹周围病变显示效果的影响，并与迭代重建（混合迭代重建[AIDR3D]）和传统滤波反投影（FBP）算法进行比较。

材料与方法：本回顾性研究纳入2021年12月至2023年11月期间在我院佳能640层CT(Aquilion ONEGENESIS,Canon Medical System Corporation,Japan)接受腹部增强CT检查的30例壶腹及壶腹周围病变患者（包括壶腹癌、壶腹腺瘤、远端胆总管胆管癌、十二指肠腺癌、胰腺腺癌）。动脉期和门静脉期扫描数据分别采用DLR（body和body sharp模式）、混合迭代重建（AIDR3D）和FBP进行重建。比较四种重建图像（DLR body、DLR body sharp、AIDR3D、FBP）在病变、胆管、胰腺、十二指肠壁、肝脏等感兴趣区的信噪比（SNR）和对比噪声比（CNR），以及图像噪声（以前腹壁皮下脂肪CT值的标准差计算）。病变CNR计算公式如下：病变-胰腺CNR=（病变CT值-胰腺CT值）/脂肪SD值；病变-胆管CNR=（病变CT值-胆管CT值）/脂肪SD值；病变-十二指肠壁CNR=（病变CT值-十二指肠壁CT值）/脂肪SD值。在主观图像质量评估中，两名放射科医师根据壶腹及壶腹周围病变的显示质量对4组图像进行独立评分排序，最好：4分，第二：3分，第三：2分，第四：1分。采用Friedman检验和Dunn-Bonferroni事后检验进行统计学比较。

结果：在动脉期和门静脉期，DLR body和DLR body sharp图像在所有感兴趣区的SNR和CNR均显著高于FBP和AIDR3D图像（均P<0.05）。DLR body（动脉期：11.86±3.45，门静脉期：11.84±2.60）和DLR body sharp（动脉期：12.82±3.04，门静脉期：12.54±2.16）图像的噪声水平显著低于FBP（动脉期：28.1±6.70，门静脉期：29.09±5.46）和AIDR3D（动脉期：17.74±4.54，门静脉期：18.31±4.03）（P均<0.05）。AIDR3D图像的SNR、CNR显著高于FBP，噪声水平显著低于FBP（P均<0.05）。DLR body（动脉期：3.83[3;4]，门静脉期：3.87[3;4]）和DLR body sharp（动脉期：3.17[3;4]，门静脉期：3.13[3;4]）的主观评分均显著高于FBP（1分）和AIDR3D（2分）（P均<0.05）。DLR body与DLR body sharp图像在CNR、图像噪声及主观评分方面均无显著差异（P均>0.05）。

结论：与FBP和AIDR3D相比，DLR图像具有更低的噪声、更高的SNR和CNR，能够提供最佳的壶腹及壶腹周围病变显示效果。

Part 1：

Purpose: To investigate whether contrast-enhancement-boost (CE-boost) in combination with hybrid iterative reconstruction (Hybrid IR, also named HIR [AIDR 3D, adaptive iterative dose reduction three dimensional]) and model-based iterative reconstruction (MBIR [FIRST, forward projected model-based IR solution]) algorithms can improve the image quality of abdominal CT angiography (CTA).

Materials and methods: This retrospective study included 50 patients who underwent abdominal CTA. Both arterial and portal phases were reconstructed using three different algorithms [filtered-back projection (FBP), AIDR 3D, and FIRST] separately. CE-boost was performed additionally to generate AIDR 3D-boost and FIRST-boost images. Signal-to- noise ratio (SNR) and contrast-to-noise ratio (CNR) of the arteries and portal system were compared among the five datasets (FBP, AIDR  3D, FIRST, AIDR  3D-boost, FIRST-boost).  In subjective analyses, two radiologists independently ordered images (5, best; 1, worst) based on the visual image quality of distal arterial or portal venous branches. The Friedman and the Dunn-Bonferroni post-hoc tests were used for statistical analysis.

Results: FIRST-boost arterial and portal images had the lowest noise compared with FBP, AIDR 3D, FIRST, and AIDR 3D-boost images (all P < 0.05), and significantly higher SNR and CNR than FBP, AIDR 3D, and FIRST images (all P < 0.05). AIDR 3D-boost images showed lower noise, and higher SNR and CNR than FBP and AIDR 3D images (all P < 0.05). FIRST-boost images had higher subjective grading scores than FBP, AIDR 3D, and AIDR 3D-boost images (all P < 0.05).

Conclusion: The postprocessing technique CE-boost can improve the image quality of abdominal CTA images. MBIR in combination with CE-boost (FIRST-boost) images had the best image quality compared with the other four image datasets.

Part 2：

Purpose: To evaluate the effects of a deep learning reconstruction (DLR) method advanced intelligent clear IQ engine on the clear visibility of the biliary system in patients with dilatation of extrahepatic bile duct on contrast-enhanced(CE) CT images, in comparison with different iterative reconstruction (IR) algorithms including the adaptive iterative dose reduction 3D (AIDR 3D) algorithm, forward projected model-based IR（FIRST） as well as conventional filtered back projection (FBP) algorithm.

Materials and methods: A total of 30 patients subjected to abdominal CE-CT and found with dilatation of the extrahepatic bile duct were retrospectively included. Images of the portal phase were reconstructed using four different algorithms (FBP, AIDR 3D, FIRST, and DLR). Ten patients were with measurable bile duct lesions. Signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of the dilated bile duct, liver parenchyma, measurable bile duct lesions, and image noise were compared in the four datasets. In subjective analyses, two radiologists independently scored the image quality of the four datasets with the best (score 4), second (score 3), third (score 2), and fourth (score 1) based on the noise and visual image quality of the biliary system. The Friedman and the Bonferroni—Dunn post-hoc tests were used for comparison.

Result: DLR images yielded significant higher CNR (bile duct: 4.42±0.87, liver parenchyma: 3.78±1.47) than FBP (bile duct: 2.21±1.02, liver parenchyma: 1.43±1.29), AIDR 3D (bile duct: 2.81±0.91, liver parenchyma: 2.39±1.94) and FIRST (bile duct: 2.51±1.24, liver parenchyma: 2.45±1.81) images (all P<0.05), higher SNR (bile duct: 1.39±0.85, liver parenchyma: 9.75±1.90 ) than FBP (bile duct: 0.86±0.63, liver parenchyma: 3.31±1.12) and FIRST (bile duct: 1.01±0.61, liver parenchyma: 5.73±1.37) images (all P<0.05), and showed lower image noise (10.51±3.53) than FBP (24.10±3.92), AIDR 3D (15.72±2.41) and FIRST (17.20±3.82) images (all P<0.05). There were no significant differences in SNR and CNR between FIRST and AIDR 3D images (all P>0.05). DLR images obtained higher score (4) than FPB (1), AIDR3D [2.85 (2;3)]and FIRST [2.15 (2;3)] (all P<0.05).

Conclusion: Deep learning reconstruction algorithm improved subjective and objective image quality of the biliary system on abdominal contrast-enhanced CT. 

Part 3：

Purpose: To evaluate the effect of a deep learning reconstruction (DLR) method on the conspicuity of ampullary and periampullary lesions on abdominal CT image, in comparison with iterative reconstruction (IR) algorithms (hybrid IR [AIDR3D]) and conventional filtered back projection (FBP). 

Materials and methods: This retrospective study included 30 patients who underwent contrast-enhanced abdominal imaging at our hospital between December 2021 and November 2023 with ampullary and periampullary lesions (include ampullary cancer, ampullary adenoma, distal common bile duct cholangiocarcinoma, duodenal adenocarcinoma, pancreatic adenocarcinoma). The arterial and portal venous phase scan data were reconstructed to obtain DLR (body and body sharp), HIR (AIDR3D) and FBP. Signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of the lesions, bile duct, pancreas, duodenum wall, liver, as well as image noise (which was calculated as the SD of the CT values for anterior subcutaneous fat) were compared among the four datasets (DLR body, DLR body sharp, AIDR3D, FBP）. CNR of the lesions were calculated as follow: CNR lesion to pancreas = (HU[lesion] – HU[pancreas])/SD[fat]，CNR lesion to bile duct  = (HU[lesion] – HU[common bile duct])/SD[fat], CNR lesion to duodenum wall  = (HU[lesion] – HU[duodenum wall])/SD[fat]. In qualitative image analyses, two radiologists independently ordered images (best: 4, worst: 1) based on the visual image quality of ampullary and periampullary lesions. The Friedman and the Dunn-Bonferroni post-hoc tests were used for comparison. 

Results: DLR body and DLR body sharp images yielded significant higher SNR and CNR than FBP and AIDR3D images in all regions of interest in both arterial and portal venous phases (all P<0.05). DLR body（arterial : 11.86±3.45, portal venous : 11.84±2.60）and DLR body sharp（12.82±3.04, 12.54±2.16）images showed significant lower image noise than FBP (28.1±6.70, 29.09±5.46) and AIDR3D (17.74±4.54, 18.31±4.03) in both arterial and portal venous phases (all P<0.05). AIDR3D showed significant higher SNR and CNR, and lower image noise than FBP images (all P<0.05). DLR body (arterial: 3.83[3;4], portal venous: 3.87[3;4]) and DLR body sharp (3.17[3;4], 3.13 [3;4]) images obtained higher score than FPB (1,1) and AIDR3D (2,2) (all P<0.05). There were no significant differences in CNR, image noise, as well as qualitative scores between DLR body and DLR body sharp images (all P>0.05).