Biomedical Engineering Reference
In-Depth Information
Table 1 Age (range and mean), BMI (range and mean) of the study population at baseline,
rescanned (one week later from baseline visit) and also number of knees for the distribution of KL
are elaborated
Baseline
Rescanned
(288 knees)
(31 knees)
Age (years)
21-81 (56)
26-75 (61)
BMI (kg/m
2
)
19-38 (27)
21-34 (27)
KL 0
145
11
KL 1
88
13
KL 2
31
2
KL 3
23
5
KL 4
1
0
where
CI
p
and
CI
q
are the congruity indices at the locations
p
and
q
respectively.
The quantification method involves few parameters that are curvature scale, step
size, and number of iterations in the mean curvature flow. The step size was fixed
at 0.15. The optimized parameters were the scale at which the curvatures were
computed and the number of iterations in the mean curvature flow.
2.5 Study Population and Image Acquisition
We validated the CI quantification on a study population consisting of 159 subjects.
The Radiographs were acquired for both the knees simultaneously using a X-ray
scanner. The MRI scans were acquired using an Esaote C-span 0.18T scanner with
the subjects in the supine, nonweight bearing position. The scanner parameters
were as follows: 40
flip angle, 50-ms repetition time, and 16-ms echo time. The in-
plane resolution was 0.7 mm x 0.7 mm and slice thickness between 0.7 mm and
0.9 mm. The severity of the OA was graded from radiographs based on Kellgren
and Lawrence index (KL index) [
14
]. We have 288 knee scans at baseline for
validating diagnostic ability of congruity after excluding 25 knees that were used
for training of segmentation classifier and remaining due to poor image quality. We
also have 31 knees that were rescanned one week later for validating the measure
precision. The detailed study population is presented in Table
1
.
2.6 Statistical Analysis
The precision of the quantification was computed as the Root mean squared
coefficient of variation (RMS CV). The diagnostic ability of the measurement to
separate any two groups is calculated from area under the curve (AUC) from
receiver operating characteristics (ROC). The statistical significance of the AUC
value is computed from the Delong test [
15
].
