Keywords:
Computer Applications-Detection, diagnosis, CAD, CT, Lung, Metastases
Authors:
D. V. Nesterov1, E. Rozengauz2, V. Nesterova1, Z. Alderov1; 1Saint-Petersburg/RU, 2St. Petersburg/RU
DOI:
10.1594/ecr2018/C-2895
Methods and materials
At Russian Scientific Center for Radiology and Surgical Technologies (Saint-Petersburg) database CT images of 7 patients with disseminated metastatic lung lesions were evaluated.
We have chosen 27 perivascular and parapleural nodules.
Images were segmented and analized in Mango v4.0.1 (http://ric.uthscsa.edu/mango/).
Segmentation was made with following algorithm.
-
Creation of sphere ROI with diameter 2x larger then nodule diameter .
-
Exclusion from ROI voxels with density that is different from nodule density using histogram tool Fig. 2 Fig. 3.
-
Exclusion of structures that are not connected with nodule Fig. 4.
As a measure of variability variation coefficient was chosen [1].
We evaluate several factors that can affect measurements variability: nodule size,
area of contact between nodule and other structures (vessels,
pleura),
observer experience.
The area of contact was measured with self-developed software mean ROI,
made after all corrections were made.
Three observers with experience in CT 0,
1 and 5 years three times manually correct nodule contours.
To evaluate effect of observer experience variation coefficients were compared with ANOVA test.
Effect of size and contact area was measured with linear model analisis.
All statistics was made with R.