Skip to main content

Table 3 Qualitative Image Quality Assessment Results

From: 75% radiation dose reduction using deep learning reconstruction on low-dose chest CT

 

QLD-DLIR

LD-IR

P-value*

ICC

[95% CI]

Lung-window setting

    

Subjective noise

3.23 ± 0.19

(3, 3.33)

3.27 ± 0.22

(3, 3.33)

0.11

0.26

[0.12, 0.38]

Spatial resolution

3.14 ± 0.28

(3, 3.33)

3.16 ± 0.27

(3, 3.33)

0.12

0.51

[0.42, 0.59]

Distortion artifact

3.08 ± 0.24

(3, 3.33)

3.07 ± 0.19

(3, 3)

0.78

0.52

[0.43, 0.60]

Mediastinal-window setting

    

Subjective noise

3.00 ± 0.19

(3, 3)

3.05 ± 0.14

(3, 3)

0.045

0.88

[0.85, 0.87]

Spatial resolution

2.92 ± 0.29

(2.67, 3)

3.22 ± 0.29

(3, 3.33)

< 0.001

0.39

[0.28, 0.49]

Distortion artifact

2.75 ± 0.18

(2.67, 2.92)

3.04 ± 0.14

(3, 3)

< 0.001

-0.34

[-0.58, -0.13]

Beam-hardening artifact

3.01 ± 0.35

(2.75, 3.33)

3.47 ± 0.27

(3.33, 3.67)

< 0.001

0.64

[0.57, 0.69]

Overall image quality

3.14 ± 0.21

(3, 3.33)

3.17 ± 0.17

(3, 3.33)

0.15

0.45

[0.35, 0.54]

  1. Note.—Results from 3-mm, standard-kernel images are presented. Data are presented as means ± standard deviations (interquartile range). Higher scores indicate better image quality. Italicized data indicate that the values are higher than the compared counterpart. CI = confidence interval, DLIR = deep-learning image reconstruction, ICC = intraclass correlation coefficient, IR = iterative reconstruction, LD = low dose, QLD = quarter of the low dose
  2. *P-values were calculated using the paired t-test or Wilcoxon signed-rank test, as appropriate