PART 01

英文摘要

Low-dose abdominal CT with deep learning reconstruction achieves image quality superior to standard-dose: a dual clinical and phantom validation

Purpose

To evaluate image quality improvement in low-dose abdominal CT using different weighting of a deep learning reconstruction algorithm.

Methods and Materials

A dual-validation study (clinical and phantom) was performed on a 256-row, 16-cm wide-detector CT. The clinical study enrolled 60 patients undergoing arterial phase abdominal CT. Group A (control, n=30) received standard dose (CTDI: 12.25) with iterative reconstruction (ID, ID1). Group B (low dose, CTDI: 8.74) was reconstructed with ID1, VC1, VC2, and VC3. The phantom study used iodine contrast (2.5 mg/mL) to simulate arterial enhancement, with control (group C) and experimental (group D) groups matching clinical settings. CT values and SD of the liver, spleen, aorta, paraspinal muscle, and phantom holes were measured; SNR was calculated; and paired t-tests were used. Two radiologists blindly assessed overall image quality using a Likert scale.

Results

No significant differences in BMI or CT values were found between groups A and B (P>0.05). CTDI values were 12.25±0.56 vs. 8.74±0.58. Compared with standard-dose A-ID1, low-dose B-ID1 showed significantly degraded image quality. In group B, higher VC weighting reduced SD and increased SNR. B-VC3 achieved lower noise and better image quality than A-ID1 (P<0.05). For A-ID1: liver CT/SD = 66.74±5.18 HU / 14.13±2.99 HU; paraspinal muscle = 56.92±3.95 HU / 14.71±3.31 HU. For B-VC3: liver = 66.09±7.07 HU / 9.42±1.69 HU; paraspinal = 58.40±5.55 HU / 9.31±1.73 HU (P<0.05). Phantom results were consistent: D-VC3 performed best (P<0.05). For iodine contrast (2.5 mg/mL), CT/SD were 72.7±1.62 HU / 8.81±1.81 HU for C-ID1 vs. 72.59±2.91 HU / 7.91±0.95 HU for D-VC3 (P<0.05). Subjective scores: B-VC3 > B-VC2 > B-VC1 > A-ID1 > B-ID1 (P<0.05); inter-observer Kappa = 0.82.

Conclusion

Low-dose arterial phase abdominal CT with deep learning reconstruction (VC) achieves image quality comparable to or better than standard-dose CT.

Clinical Relevance/Application

This study provides a solid foundation for low-dose clinical implementation by demonstrating that low-dose arterial phase abdominal CT with VC can match or exceed standard-dose image quality.

PART 02

图表与图注

PART 03

中文摘要

低剂量腹部CT结合深度学习重建获得优于标准剂量的图像质量：临床与体模双重验证

目的

评价不同权重深度学习重建算法对低剂量腹部CT图像质量的改善作用。

材料与方法

在256排、16 cm宽探测器CT上开展临床与体模双重验证研究。临床研究纳入60例接受腹部动脉期CT检查的患者。A组（对照组，n=30）采用标准剂量（CTDI：12.25）扫描和迭代重建（ID，ID1）；B组采用低剂量（CTDI：8.74）扫描，分别使用ID1、VC1、VC2和VC3重建。体模研究采用2.5 mg/mL碘对比剂模拟动脉期强化，设置与临床条件相匹配的对照组（C组）和实验组（D组）。测量肝脏、脾脏、主动脉、椎旁肌及体模孔洞的CT值和标准差（SD），计算信噪比（SNR），并采用配对t检验。两名放射科医师使用Likert量表，对整体图像质量进行盲法评价。

结果

A、B两组的BMI和CT值差异无统计学意义（P＞0.05）。CTDI分别为12.25±0.56和8.74±0.58。与标准剂量A-ID1相比，低剂量B-ID1的图像质量明显下降。在B组中，随着VC权重增加，SD下降、SNR提高；B-VC3的噪声更低，图像质量优于A-ID1（P＜0.05）。A-ID1中，肝脏CT值/SD为66.74±5.18 HU / 14.13±2.99 HU，椎旁肌为56.92±3.95 HU / 14.71±3.31 HU；B-VC3中，肝脏为66.09±7.07 HU / 9.42±1.69 HU，椎旁肌为58.40±5.55 HU / 9.31±1.73 HU（P＜0.05）。体模结果与临床结果一致，D-VC3表现最佳（P＜0.05）。对于2.5 mg/mL碘对比剂，C-ID1的CT值/SD为72.7±1.62 HU / 8.81±1.81 HU，D-VC3为72.59±2.91 HU / 7.91±0.95 HU（P＜0.05）。主观评分排序为B-VC3＞B-VC2＞B-VC1＞A-ID1＞B-ID1（P＜0.05）；观察者间一致性Kappa=0.82。

结论

低剂量腹部动脉期CT结合深度学习重建（VC）可获得与标准剂量CT相当或更优的图像质量。

临床相关性

本研究表明，低剂量腹部动脉期CT结合VC能够达到或超过标准剂量图像质量，为低剂量方案的临床应用提供依据。