This poster is published under an
open license. Please read the
disclaimer for further details.
Keywords:
Neuroradiology brain, Computer applications, MR, Computer Applications-Detection, diagnosis, Comparative studies
Authors:
K. Schwerdtner1, M. Domin1, H.-J. Huppertz2, M. Grothe1, E. Rathmann1, S. Langner1; 1Greifswald/DE, 2Zurich/CH
DOI:
10.1594/ecr2017/C-3044
Results
Results - Patients
During the study period,
357 patients with clinically proven or suspected chronic demyelinating central nervous disease underwent baseline MRI and at least one follow-up study.
- 77 patients were excluded due to MR imaging protocol violations.
- 9 patients were excluded due to incomplete segmentation process.
- In 16 patients the diagnosis of a chronic demyelinating central nervous disease could not be confirmed.
255 patients were included in the final evaluation.
85 were male (33.3%) and 170 female (66.7%),
the mean age was 45.1 years (range = 18.0 – 77.8).
Results – MR imaging
255 patients had 804 MRI examinations (median = 3; mean = 3.15; range = 1 – 8).
From these 804 examinations 550 subtraction datasets could be generated (median = 2; mean = 2.16; range= 1 – 7).
19 subtraction datasets had to be excluded due to severe imaging artifacts and in 4 patients the written report was unavailable.
Final analysis comprised 531 subtraction datasets.
Results - Determination of changes in T2 Lesion Load and comparison values
sadcT2ll found a change 164 of 531 SI (30.89%),
while radiological assessment found a change in 193 of 531 SI (36.35%).
Results of arbitration
After establishing the standard with arbitration as mentioned above cT2ll were found in 192 SI (36.16%).
The were following results of a detected change regardless to its correct quantification:
- In 163 cases the result of sadcT2ll was TP compared to 178 cases of racT2ll,
respectively.
- Results of sadcT2ll were FP in 1 cases,
racT2ll in 8 cases,
respectively.
- TN results occurred 338 times in sadcT2ll and 331 times in racT2ll,
respectively.
- In 29 cases sadcT2ll was FN and in 14 cases racT2ll,
respectively.
|
true positive
|
false positive
|
true negative
|
false negative
|
sadcT2ll
|
163
|
1
|
338
|
29
|
racT2ll
|
178
|
8
|
331
|
14
|
Table 1: Results for detection of cT2ll regardless of correct quantification
Regarding the correct assessment of the number of cT2ll there were following results:
- In 133 cases the result of sadcT2ll was TP compared to 150 cases of racT2ll,
respectively.
- Results of sadcT2ll were FP in 2 cases,
racT2ll in 15 cases,
respectively.
- TN results occurred 338 times in sadcT2ll and 331 times in racT2ll,
respectively.
- In 58 cases sadcT2ll was FN and in 35 cases racT2ll,
respectively.
|
true positive
|
false positive
|
true negative
|
false negative
|
sadcT2ll
|
133
|
2
|
338
|
58
|
racT2ll
|
150
|
15
|
331
|
35
|
Table 2: Results for quantitative detection of cT2ll
Results - Statistics
The following table shows the results of both methods to detect changes of T2 lesion load regardless of its correct quantification.
|
sadcT2ll
|
racT2ll
|
sensitivity
|
84.9%
|
92.7%
|
specificity
|
99.7%
|
97.6%
|
PPV
|
99.4%
|
95.7%
|
NPV
|
92.1%
|
95.9%
|
FPR
|
0.3%
|
2.4%
|
FNR
|
15.1%
|
7.3%
|
LH+
|
287.80
|
39.29
|
LH-
|
0.15
|
0.07
|
Table 3
The following table shows the results of both methods to detect quantitatively correct changes of T2 lesion load:
|
sadcT2ll
|
racT2ll
|
sensitivity
|
69.6%
|
81.1%
|
specificity
|
99.4%
|
95.7%
|
PPV
|
98.5%
|
90.9%
|
NPV
|
85.4%
|
90.4%
|
FPR
|
0.6%
|
4.3%
|
FNR
|
30.4%
|
18.9%
|
LH+
|
118.38
|
18.70
|
LH-
|
0.31
|
0.20
|
Table 4