Scientific Papers

Computer-assisted evaluation of small airway disease in CT scans of Iran-Iraq war victims of chemical warfare by a locally developed software: comparison between different quantitative methods | BMC Medical Imaging


In this study, 73 men participated, including 46 (63.0%) in the patient group and 27 (37.0%) in the control group. The participants’ baseline characteristics are described in Table 1.

Table 1 Baseline participant’s characteristics

Association of QCT measurements with spirometry results

Table 2 shows the association of all QCT measurements with spirometry parameters. There was a significant negative correlation between PRMFsad and Fev1/Fvc (r = -0.610, P-value < 0.001). Moreover, a significant negative correlation was found between ATI and Fev1/Fvc (r = -0.602, P-value < 0.001). The best correlation was observed between %LAAExp < -856 and Fev1/Fvc (r = -0.665, P-value < 0.001). Besides, PRMFsad + PRMEmph demonstrated a strong negative correlation with Fev1/Fvc (r = -0.617, P-value < 0.001). The E/I ratio had a moderate negative correlation with Fev1/Fvc and MEEF (r = -0.429 and r = -0.323, respectively, with P-value < 0.01). The results revealed a significant negative correlation between PRMEmph and Fev1/Fvc (r = -0.445, P-value < 0.001). In addition, there was a significant negative correlation between %LAAins < -950 and Fev1/Fvc (r = -0.506, P-value < 0.001; Table 2).

Table 2 Pearson correlation coefficient between QCT measurements and Spirometry parameters for both groups of study

Univariate linear regression analysis confirmed %LAAExp < -856 as a significant predictor of Fev1/Fvc (β %LAAExp<-856 = -96.02, P-value = 0.002). Also, univariate linear regression analysis confirmed that ATI and %LAAExp < -856 were significant predictors of MMEF (βATI = 224.73, P-value = 0.033, β %LAAExp<-856 = -253.23, P-value = 0.007, respectively).

Association of air trapping indices with emphysema

To illustrate the association between PRMEmph and air trapping, a scatter plot and linear correlation coefficient were used. All the graphs showed a straight line with a positive slope. There was a positive and moderate relationship between PRMEmph and ATI (r = 0.417, P-value < 0.001). Significant relationships were also observed between PRMEmph and %LAAExp < -856 and between PRMEmph and PRMFsad (r = 0.641, P-value < 0.001, r = 0.396, P-value = 0.001, respectively). All the mentioned correlations in any pair were statistically different from each other (P-value < 0.001; Fig. 2).

Fig. 2
figure 2

Scatter plot of PRMEmph with three classes of air trapping (PRMFsad -ATI-LAA < -856)

The scatter plot in Fig. 3 displays the significant positive relationship between PRMEmph and E/I ratio (r = 0.373, P-value = 0.001).

Fig. 3
figure 3

Scatter plot of PRMEmph with the E/I ratio

Similarly, the Pearson correlation test was applied to assess the association between %LAAins < -950, as a measure of emphysema, and three classes of air trapping (PRMFsad, ATI, and %LAAExp < -856). The correlation between %LAAins < -950 and %LAAExp < -856 was significant and positive (r = 0.526, P-value < 0.001). In addition, %LAAins < -950 significantly correlated with PRMFsad and ATI (r = 0.373, P-value = 0.001, r = 0.462, P-value < 0.001, respectively).

Comparison of three classes of air trapping measurement

Graphs and correlation tests were used to compare the correlation statistics with Fev1/Fvc across three classes of AT measurement (Fig. 4).

Fig. 4
figure 4

Scatter plots of three AT measurement methods with Fev1/Fvc: Correlation between %LAAExp < -856 and Fev1/Fvc (A; R2 = 0.442), Correlation between PRMFsad and Fev1/Fvc (B; R2 = 0.372), Correlation between ATI and Fev1/Fvc (C; R2 = 0.363)

Table 2 presents the significant negative correlation between the three classes of AT measurement and Fev1/Fvc. There were significant differences in correlation coefficients for the pairs of %LAAExp < -856-Fev1/Fvc and PRMFsad-Fev1/Fvc (r = -0.665 and r = -0.610, respectively; P-value = 0.043) and for the pairs of %LAAExp < -856-Fev1/Fvc and ATI-Fev1/Fvc (r = -0.665, r = -0.602, respectively; P-value = 0.010). However, there was no significant difference in the correlation coefficient for the pairs of PRMFsad-Fev1/Fvc and ATI-Fev1/Fvc (r = -0.544, r = -0.606, respectively; P-value = 0.385).

Table 2 indicates the significant negative correlation between three classes of AT measurement and MMEF, but there were no significant differences in correlation coefficients for the pairs of %LAAExp < -856-MMEF and PRMFsad-MMEF (r = -0.432, r = -0.388, respectively; P-value = 0.125) and the pairs of PRMFsad-MMEF and ATI-MMEF (r = -0.388, r = -0.351, respectively; P-value = 0.125). However, there was a significant difference in the correlation coefficient for the pairs of %LAAExp < -856-MMEF and ATI-MMEF (r = -0.432, r = -0.351, respectively; P-value = 0.006).

Comparison of QCT measurements between the two groups

Table 3 compares QCT measurements between the two groups. The t-test results showed that for conventional methods, only the E/I ratio was statistically different between the two groups (P-value < 0.001). Moreover, PRMEmph significantly differed between the patient and control groups (P-value < 0.001). All the AT measurements (PRMFsad, %LAAExp < -856, and ATI) significantly differed between the case and control groups (P-value < 0.001; Table 3).

Table 3 Comparisons of QCT measurements between case and control groups

A binary logistic regression model was applied to correlate PRMFsad with the likelihood of a participant being in the patient group. Age and sex were matched in both groups because all the participants were male, and the t-test showed no significant difference in age between the patient and control groups (P-value = 0.577). The logistic regression model demonstrated the significant effect of PRMFsad after adjusting for emphysema as a confounder (ORadj = 1.30, P-value = 0.001), so PRMFsad was approved as a significant predictor of the outcome (being a patient). Then, to find an optimal cut-point to classify the participants into the case and control groups, the ROC curve analysis was applied using PRMFsad as an independent variable (Fig. 5a and Table 4). PRMFsad significantly identified patients with an area under the ROC curve of 0.80 (P-value < 0.001; Table 4). ROC analysis generated an optimal PRMFsad cut-point of 19% of the total lung volume. Values equal to or greater than 19% of PRMFsad identified patients with a sensitivity of 0.78 and specificity of 0.70.

Fig. 5
figure 5

a ROC curve to illustrate diagnostic ability of logistic regression model, based on PRMFsad as a predictor. b ROC curve to illustrate the diagnostic ability of the logistic regression model, based on %LAAExp < -856 as a predictor. c ROC curve to illustrate diagnostic ability of logistic regression model, based on ATI as a predictor

Table 4 Area under ROC curve

The binary logistic regression model also showed the significant impact of %LAAExp < -856 on the outcome, adjusted for emphysema as a confounder (ORadj = 1.18, P-value = 0.001). Next, ROC analysis using %LAAExp < -856 as a strong predictor was performed to find an optimal cut-point for classifying the participants into the two groups (Fig. 5b).

Table 4 (The second row) shows the significant and valuable accuracy of the logistic model based on %LAAExp < -856 as a significant predictor (AUC = 0.79, P-value < 0.001). ROC analysis identified the value of 0.23 as an %LAAExp < -856 optimal cut-point with a sensitivity of 0.72 and specificity of 0.70. The %LAAExp < -856 value > 0.23 of the total lung volume assigned the participants to the patient group.

Similarly, the binary logistic regression model expressed the significant impact of ATI on the outcome, adjusted for emphysema as a confounder (ORadj = 1.16, P-value = 0.001). Subsequently, ROC analysis using ATI as a relatively strong predictor was performed to find an optimal cut-point to classify the participants into two groups (Fig. 5c).

Table 4 (The third row) lists the significant and valuable accuracy of the logistic model based on ATI as a significant predictor (AUC = 0.78, P-value < 0.001). ROC analysis identified the value of 0.27 as an ATI optimal cut-point with a sensitivity of 0.70 and specificity of 0.70. The ATI value > 0.27 of lung volume with radio-density of -856 to -950 HU in inhaled CT assigned the participants into the patient group.

The ability of PRMFsad, ATI, and %LAAExp < -856 to characterize AT is demonstrated in representative CT images in a patient and a control participant (Fig. 6).

Fig. 6
figure 6

AT maps from A) static threshold of -856HU, B) the Parametric Response Map(PRM), and C) the Air Trapping Index(ATI) method in a patient (lower row), and control subject (upper row)



Source link