Skip to main content

Table 5 Qualitative analysis

From: Image quality comparison of lower extremity CTA between CT routine reconstruction algorithms and deep learning reconstruction

Parameter

DLR

MBIR

HIR

FBP

p

Aorta

     

Noise

4.98 [4.91–5.05]

3.94 [3.84–4.03]*

4.08 [4.00–4.16]*

3.19 [3.07–3.30]*°˜

 < 0.001

Subjective sharpness

4.94 [4.87–5.01]

4.56 [4.41–4.71]*

4.23 [4.11–4.35]*

3.69 [3.55–3.82]*°˜

 < 0.001

Natural appearance

4.23 [4.09–4.37]

3.95 [3.88–4.02]

4.00 [3.89–4.10]

3.85 [3.73–3.96]

 < 0.001

Femoral artery

     

Noise

4.90 [4.81–4.99]

3.92 [3.80–4.03]*

4.06 [3.97–4.16]*

3.00 [2.94–3.06]*°˜

 < 0.001

Subjective sharpness

4.94 [4.87–5.00]

4.52 [4.37–4.67]*

4.14 [4.04–4.25]*

3.50 [3.35–3.65]*°˜

 < 0.001

Natural appearance

4.28 [4.13–4.43]

4.00 [3.93–4.07]

3.97 [3.88–4.07]

3.77 [3.63–3.91]*

 < 0.001

Popliteal artery

     

Noise

4.85 [4.75–4.96]

3.90 [3.76–4.03]*

4.06 [3.97–4.16]*

3.10 [3.00–3.21]*°˜

 < 0.001

Subjective sharpness

4.79 [4.67–4.91]

4.56 [4.42–4.71]

4.08 [3.98–4.18]*°

3.73 [3.60–3.86]*°

 < 0.001

Natural appearance

4.08 [3.99–4.16]

3.97 [3.92–4.03]

3.95 [3.84–4.05]

3.87 [3.76–3.98]

0.031

Liver

     

Noise

4.81 [4.70–4.93]

3.02 [2.97–3.06]*

3.98 [3.94–4.02]*°

3.02 [2.98–3.06]*

 < 0.001

Subjective sharpness

4.60 [4.46–4.74]

3.98 [3.85–4.11]*

4.08 [4.00–4.16]*

3.15 [3.04–3.25]*°˜

 < 0.001

Natural appearance

4.67 [4.51–4.82]

3.28 [3.13–3.43]*

4.00 [3.93–4.07]*°

3.62 [3.45–3.78]*

 < 0.001

Psoas muscle

     

Noise

4.90 [4.81–4.99]

3.1 [3.00–3.21]*

4.00 [3.92–4.08]*°

2.94 [2.87–3.01]*

 < 0.001

Subjective sharpness

4.92 [4.84–5.00]

4.15 [4.03–4.27]*

4.08 [4.00–4.16]*

3.1 [3.01–3.19]*°˜

 < 0.001

Natural appearance

4.46 [4.30–4.62]

3.46 [3.30–3.63]*

3.97 [3.88–4.07]°

3.67 [3.51–3.82]*

 < 0.001

  1. Estimated marginal means for every quality parameter, reconstruction algorithm and pairwise comparison significance are given, with 95% confidence intervals in brackets
  2. DLR deep learning–based reconstruction, MBIR model-based iterative reconstruction, HIR hybrid-iterative reconstruction, FBP filtered back projection
  3. Pairwise comparisons showed significant differences from DLR (*), MBIR (°), and HIR (˜) (p < 0.05)