Keywords:
Abdomen, Emergency, Gastrointestinal tract, CT, Image manipulation / Reconstruction, Diagnostic procedure, Technology assessment, Medico-legal issues, Foreign bodies, Forensics
Authors:
J. Rey, V. Mino, A.-S. KNOEPFLI, J.-Y. Meuwly, F. R. Verdun, R. A. MEULI, S. Schmidt; Lausanne/CH
DOI:
10.1594/ecr2012/C-1138
Results
The increase of ASIR technique significantly diminished the objective noise measured on the images (p<0.001).
The analysis of adrenal gland (p=0.206) and the identification of the common bile duct (p=0.450) did not depend on the type of image reconstruction (FBP,
30% ASIR or 60% ASIR),
neither did the time of reading the examinations (p=0.901)
The overall sensitivity and specificity of all four readers for the detection of cocaine-filled packets (without taking account of the exact number) was 91.1% and 84.8%, respectively (Fig.
3,
Fig.
4). The area Az under the ROC curve was 0.92.
The individual results can be seen in Table 1,
the result obtained for the differently reconstructed images (FBP,
30% ASIR,
60% ASIR) is demonstrated in Table 2.
Table 1
Diagnostic value of the 4 readers concerning the detection (absence/presence) of cocaine-filled packets in the digestive tract
|
|
Sensitivity (%)
|
Specificity (%)
|
Area Az (ROC curve)
|
L 1
|
100
|
100
|
1
|
L 2
|
90.7
|
90.7
|
0.953
|
L 3
|
88.9
|
75.2
|
0.876
|
L 4
|
85.0
|
74.4
|
0.871
|
Table 2
Diagnostic value of the three types of reconstructed images concerning the detection (absence/presence) of cocaine-filled packets in the digestive tract
|
|
Sensitivity (%)
|
Specificity (%)
|
Area Az (ROC curve)
|
FBP
|
92.3
|
86.1
|
0.930
|
30% ASIR
|
90.6
|
84.2
|
0.921
|
60% ASIR
|
90.6
|
84.2
|
0.921
|
FBP = filtered-back projection,
ASIR = adaptive statistical iterative reconstructions
There was no significant difference in sensitivity or specificity between the three types of reconstructed images (p=0.333,
Table 2).
The readers’ diagnostic certitude, estimated when reading the images,
did not significantly vary between the three types of reconstructed images (p=.0.891).
Concerning the number of cocaine-filled packets located in the digestive tract,
model-based linear regression showed a prediction of 93.35% for detecting the correct number of CFP.
The interobserver agreement kappa between the four readers to detect the correct number and position of the CFP varied from 0.51 to 0.79 (mean = 0.62), thus considered as good according to Fleiss (8).