13 Modeling and Prediction of Choroidal Neovascularization …
369
In this chapter, we introduce a CNV growth model for longitudinal OCT images
based on the reaction diffusion model [10]. Finite element method (FEM) is used to
solve the model equation. Optimal growth parameters are obtained by minimizing an
objective function measuring prediction accuracy. The method was tested on a dataset
with 7 patients, each with 12 monthly-scanned OCT images. The experimental results
showed the accuracy of the proposed method.
13.2 Method
13.2.1 Method Overview
Figure 13.2 shows the framework of our method. Suppose there are N longitudinal
images for each subject in the study. The first N − 1 images are used for training,
from which the growth parameters are obtained to predict the Nth image. First,
image preprocessing, including registration and segmentation, are conducted on all
OCT images. Secondly, tetrahedral meshes are constructed for the segmented CNV
volumes and the related retinal regions. Thirdly, the CNV growth model is applied
on the first N − 1 images to get their growth parameters are learned by optimization.
Then, the growth parameter for the Nth image is estimated from the previous ones
by curve fitting, and is used in the model to obtain the predicted Nth image. The
prediction result is validated by comparing the synthesized image with the ground
truth of the Nth image.
13.2.2 Data Acquisition
Seven eyes from seven subjects diagnosed with CNV associated with AMD were
scanned once a month during a one year period. 3D OCT images with 512 × 128 ×
1024 voxels (each with size of 11.72 × 46.88 × 1.95 µm
3 ), covering the volume of
6 × 6 × 2 mm
3 were obtained by Zeiss Cirrus OCT scanner (Carl Zeiss Meditec, Inc.,
Dublin, CA). The subjects were enrolled in a trial of anti-VEGF medicine. In this trial,
patients were randomly divided into the treatment group or the reference group. The
treatment plan included two phases: the core treatment and the extended treatment.
The core treatment involved 3 monthly intravitreal anti-VEGF injections (conbercept,
0.5 mg), and the extended treatment meant injections in three-month intervals. The
difference between the two groups was that: the reference group started the real
treatment 3 months later than the treatment group, and was given condolences agent
in the first 3 month. Figure 13.3 shows the detailed treatment plans of treatment group
and reference group, respectively. Among the 7 subjects studied in this experiment,
4 subjects were in the treatment group and the rest 3 were in the reference group.
369
In this chapter, we introduce a CNV growth model for longitudinal OCT images
based on the reaction diffusion model [10]. Finite element method (FEM) is used to
solve the model equation. Optimal growth parameters are obtained by minimizing an
objective function measuring prediction accuracy. The method was tested on a dataset
with 7 patients, each with 12 monthly-scanned OCT images. The experimental results
showed the accuracy of the proposed method.
13.2 Method
13.2.1 Method Overview
Figure 13.2 shows the framework of our method. Suppose there are N longitudinal
images for each subject in the study. The first N − 1 images are used for training,
from which the growth parameters are obtained to predict the Nth image. First,
image preprocessing, including registration and segmentation, are conducted on all
OCT images. Secondly, tetrahedral meshes are constructed for the segmented CNV
volumes and the related retinal regions. Thirdly, the CNV growth model is applied
on the first N − 1 images to get their growth parameters are learned by optimization.
Then, the growth parameter for the Nth image is estimated from the previous ones
by curve fitting, and is used in the model to obtain the predicted Nth image. The
prediction result is validated by comparing the synthesized image with the ground
truth of the Nth image.
13.2.2 Data Acquisition
Seven eyes from seven subjects diagnosed with CNV associated with AMD were
scanned once a month during a one year period. 3D OCT images with 512 × 128 ×
1024 voxels (each with size of 11.72 × 46.88 × 1.95 µm
3 ), covering the volume of
6 × 6 × 2 mm
3 were obtained by Zeiss Cirrus OCT scanner (Carl Zeiss Meditec, Inc.,
Dublin, CA). The subjects were enrolled in a trial of anti-VEGF medicine. In this trial,
patients were randomly divided into the treatment group or the reference group. The
treatment plan included two phases: the core treatment and the extended treatment.
The core treatment involved 3 monthly intravitreal anti-VEGF injections (conbercept,
0.5 mg), and the extended treatment meant injections in three-month intervals. The
difference between the two groups was that: the reference group started the real
treatment 3 months later than the treatment group, and was given condolences agent
in the first 3 month. Figure 13.3 shows the detailed treatment plans of treatment group
and reference group, respectively. Among the 7 subjects studied in this experiment,
4 subjects were in the treatment group and the rest 3 were in the reference group.