Statistical analysis
Diagnostic performance was evaluated by sensitivity, specificity, and
overall accuracy, with 95% confidence intervals (CIs) calculated based
on exact binomial distributions (due to sparse data in certain
categories). To compare two modalities on a specific metric, the McNemar
exact test was used. All analyses were performed in R version 4.0.2 (R
Foundation for Statistical Computing, Vienna, Austria).
Per-lesion analyses
The following summarizes the diagnostic performance of T2, DWI,
T2-DWI, PET, T2-DWI-PET by sensitivity (95% CIs) and performs pairwise
tests between the 5 (or 3) modalities (10 or 3 pairs).
Lesion-level focal
Sensitivity
Table 1a. Sensitivity (95% CI) for lesion-level focal
lesions.
T2 |
0.919 (0.859, 0.959) |
DWI |
0.667 (0.58, 0.745) |
T2-DWI |
0.933 (0.877, 0.969) |
PET |
0.444 (0.359, 0.532) |
T2-DWI-PET |
0.985 (0.948, 0.998) |
Table 1b. P-values (McNemar exact test) for pairwise comparison
of diagnostic performance on lesion-level focal lesions.
T2 vs DWI |
<0.001 |
T2 vs T2-DWI |
0.5 |
T2 vs PET |
<0.001 |
T2 vs T2-DWI-PET |
0.004 |
DWI vs T2-DWI |
<0.001 |
DWI vs PET |
<0.001 |
DWI vs T2-DWI-PET |
<0.001 |
T2-DWI vs PET |
<0.001 |
T2-DWI vs T2-DWI-PET |
0.016 |
PET vs T2-DWI-PET |
<0.001 |
False discovery rate (one minus PPV)
Between-modality tests of false discovery rate are based on the
method of Leisenring, Alono, and Pepe
(2000) using R-package DTComPair
(Stock, Hielscher, and Discacciati 2024).
## Loading required package: PropCIs
False discovery rate (false positives / all positives) and 95%
CI by modality.
0.126 (0.075, 0.204) |
0.075 (0.041, 0.132) |
0.062 (0.025, 0.15) |
P-values testing between-modality differences in false
discovery rate.
DWI v T2 |
0.043 |
DWI v PET |
0.086 |
T2 v PET |
0.723 |
Lesion-level diffuse
Table 2a. Sensitivity (95% CI) for lesion-level diffuse
lesions.
T2 |
0.846 (0.546, 0.981) |
DWI |
0.769 (0.462, 0.95) |
PET |
0.462 (0.192, 0.749) |
Table 2b. P-values (McNemar exact test) for pairwise comparison
of diagnostic performance on lesion-level diffuse lesions.
T2 vs DWI |
1 |
T2 vs PET |
0.062 |
DWI vs PET |
0.125 |
Per-patient analyses
The following summarizes the diagnostic performance of T2flex, DWI,
T2-DWI, PET, T2-DWI-PET by sensitivity, specificity, and accuracy, along
with their 95% CIs, and performs pairwise tests between the 5 modalities
(10 pairs).
Patient-level focal
Table 3a. Sensitivity, specificy, and accuracy (95% CI) for
patient-level focal lesions.
T2flex |
0.944 (0.727, 0.999) |
0.615 (0.316, 0.861) |
0.806 (0.625, 0.925) |
DWI |
0.778 (0.524, 0.936) |
0.692 (0.386, 0.909) |
0.742 (0.554, 0.881) |
T2-DWI |
0.944 (0.727, 0.999) |
0.615 (0.316, 0.861) |
0.806 (0.625, 0.925) |
PET |
0.889 (0.653, 0.986) |
0.846 (0.546, 0.981) |
0.871 (0.702, 0.964) |
T2-DWI-PET |
1 (0.815, 1) |
0.769 (0.462, 0.95) |
0.903 (0.742, 0.98) |
Table 3b. P-values (McNemar exact test) for pairwise comparison
of diagnostic performance on patient-level focal lesions.
T2flex vs DWI |
0.25 |
1 |
0.625 |
T2flex vs T2-DWI |
1 |
1 |
1 |
T2flex vs PET |
1 |
0.25 |
0.688 |
T2flex vs T2-DWI-PET |
1 |
0.5 |
0.25 |
DWI vs T2-DWI |
0.25 |
1 |
0.625 |
DWI vs PET |
0.625 |
0.5 |
0.219 |
DWI vs T2-DWI-PET |
0.125 |
1 |
0.062 |
T2-DWI vs PET |
1 |
0.25 |
0.688 |
T2-DWI vs T2-DWI-PET |
1 |
0.5 |
0.25 |
PET vs T2-DWI-PET |
0.5 |
1 |
1 |
Patient-level diffuse
Table 4a. Sensitivity, specificy, and accuracy (95% CI) for
patient-level diffuse lesions.
T2flex |
0.846 (0.546, 0.981) |
1 (0.815, 1) |
0.935 (0.786, 0.992) |
DWI |
0.769 (0.462, 0.95) |
1 (0.815, 1) |
0.903 (0.742, 0.98) |
T2-DWI |
0.846 (0.546, 0.981) |
1 (0.815, 1) |
0.935 (0.786, 0.992) |
PET |
0.462 (0.192, 0.749) |
1 (0.815, 1) |
0.774 (0.589, 0.904) |
T2-DWI-PET |
0.846 (0.546, 0.981) |
1 (0.815, 1) |
0.935 (0.786, 0.992) |
Table 4b. P-values (McNemar exact test) for pairwise comparison
of diagnostic performance on patient-level diffuse lesions.
T2flex vs DWI |
1 |
1 |
1 |
T2flex vs T2-DWI |
1 |
1 |
1 |
T2flex vs PET |
0.062 |
1 |
0.062 |
T2flex vs T2-DWI-PET |
1 |
1 |
1 |
DWI vs T2-DWI |
1 |
1 |
1 |
DWI vs PET |
0.125 |
1 |
0.125 |
DWI vs T2-DWI-PET |
1 |
1 |
1 |
T2-DWI vs PET |
0.062 |
1 |
0.062 |
T2-DWI vs T2-DWI-PET |
1 |
1 |
1 |
PET vs T2-DWI-PET |
0.062 |
1 |
0.062 |
Patient-level overall
Table 5a. Sensitivity, specificy, and accuracy (95% CI) for
patient-level overall lesions.
T2flex |
0.952 (0.762, 0.999) |
0.5 (0.187, 0.813) |
0.806 (0.625, 0.925) |
DWI |
0.857 (0.637, 0.97) |
0.6 (0.262, 0.878) |
0.774 (0.589, 0.904) |
T2-DWI |
0.952 (0.762, 0.999) |
0.5 (0.187, 0.813) |
0.806 (0.625, 0.925) |
PET |
0.857 (0.637, 0.97) |
0.8 (0.444, 0.975) |
0.839 (0.663, 0.945) |
T2-DWI-PET |
0.952 (0.762, 0.999) |
0.7 (0.348, 0.933) |
0.871 (0.702, 0.964) |
Table 5b. P-values (McNemar exact test) for pairwise comparison
of diagnostic performance on patient-level overall lesions.
T2flex vs DWI |
0.5 |
1 |
1 |
T2flex vs T2-DWI |
1 |
1 |
1 |
T2flex vs PET |
0.5 |
0.25 |
1 |
T2flex vs T2-DWI-PET |
1 |
0.5 |
0.5 |
DWI vs T2-DWI |
0.5 |
1 |
1 |
DWI vs PET |
1 |
0.5 |
0.688 |
DWI vs T2-DWI-PET |
0.5 |
1 |
0.25 |
T2-DWI vs PET |
0.5 |
0.25 |
1 |
T2-DWI vs T2-DWI-PET |
1 |
0.5 |
0.5 |
PET vs T2-DWI-PET |
0.5 |
1 |
1 |
References
Leisenring, Wendy, Todd Alono, and Margaret Sullivan Pepe. 2000.
“Comparisons of Predictive Values of Binary Medical Diagnostic
Tests for Paired Designs.” Biometrics 56 (2): 345–51.
https://doi.org/10.1111/j.0006-341x.2000.00345.x.
Stock, Christian, Thomas Hielscher, and Andrea Discacciati. 2024.
“DTComPair: Comparison of Binary
Diagnostic Tests in a Paired Study Design.”