ROC analysis of LSN (ratio)

There is a strong correlation (Spearman coefficient 0.867) between LSN score and LSN ratio (see Figure 1). So we expect their diagnostic performances to be similar.

Figure 1. Scatter plot of LSN ratio vs LSN score.

Figure 1. Scatter plot of LSN ratio vs LSN score.

The ROC curves of LSN score and LSN ratio with respect to PFI and other outcomes are plotted in Figure 2.

Figure 2. ROC curves of LSN score/ratio for various outcomes.

Figure 2. ROC curves of LSN score/ratio for various outcomes.

ROC analysis of attenuation (ratio)

The correlation between RCC attenuation and its ratio against normal attenuation is also strong (see Figure 3).

Figure 3. Scatter plot of attenuation ratio vs attenuation.

Figure 3. Scatter plot of attenuation ratio vs attenuation.

The ROC curves of attenuation and attenuation ratio with respect to PFI and other outcomes are plotted in Figure 4.

Figure 4. ROC curves of attenuation value/ratio for various outcomes.

Figure 4. ROC curves of attenuation value/ratio for various outcomes.

LSN score combined with attenuation ratio

In univariate analyses for PFI, LSN score outperforms LSN ratio (AUC: 0.639 vs 0.577) and attenuation ratio outperforms attenuation (AUC: 0.653 vs 0.636). So we combine LSN score with attenuation ratio. We fit a logistic regression of PFI against the two predictors. The ROC curves of each predictor and the model-based score are plotted in Figure 5. The combination leads to only a moderate improvement.

Figure 5. ROC curves of LSN, attenuation ratio, and a logistic model combining the two for PFI.

Figure 5. ROC curves of LSN, attenuation ratio, and a logistic model combining the two for PFI.

Subjective assessment and in combination with objective measures

The performance of subjective assessment of PFI is comparable to objective measures like LSN score and attenuation ratio (see Table 1).

Table 1. Diagnostic performance of subjective assessment of PFI.
Sensitivity (95% CI) Specificity (95% CI)
0.625 (0.475-0.775) 0.774 (0.684-0.863)

We combine subjective assessment with LSN and attenuation ratio in a multivariate logistic regression model, which leads to a moderate improvement in AUC (0.74).

Figure 6. ROC curve of logistic regression of PFI against LSN, attenuation ratio, and subjective assessment.

Figure 6. ROC curve of logistic regression of PFI against LSN, attenuation ratio, and subjective assessment.

Boxplots of LSN and attenuation

Boxplots of LSN and attenuation (ratios) are plotted below by PFI, with \(p\)-values computed from the Wilcoxon rank sum test. Their levels of significance are consistent with the AUCs, i.e., attenuation ratio > attenuation \(\approx\) LSN > LSN ratio.

Table 2. Median (inter-quartile range) of LSN and attenuation.
No PFI PFI p-value
LSN 5.5 (4.9, 6.7) 6.2 (5.6, 7.3) 0.014
LSN Ratio 1.2 (1, 1.5) 1.3 (1.1, 1.6) 0.179
Attenuation 75.9 (57.3, 92.5) 63.1 (40.6, 77.8) 0.014
Attenuation Ratio 0.8 (0.7, 1) 0.7 (0.5, 0.9) 0.006

Intra- and Inter-rater agreement of LSN

The Bland-Altman plot of the two LSN scores by Rater 1 is shown on the left panel and that between Raters 1 and 2 is shown on the right panel. The intra-class correlation coefficients are indicated in the titles.