12 Segmentation of Symptomatic Exudate-Associated …
355
the sub-graph G SS is labeled as “sink”, i.e. z-S SS (x, y) > d, then a directed arc with a
penalty weight from each node G R (x, y, z) to G SS (x, y, z + d) will be added, as shown
in Fig. 12.6b. For G R and G SI, the same approach is employed.
12.4 Performance Evaluation
12.4.1 Experimental Methods
For the initialization, we compared the performance before and after probability
normalization to show the efficacy of the probability normalization.
For the segmentation, we compared the traditional GC in [13], the traditional
GS method in [29] and the proposed probability constraints GS-GC. A multivariate
analysis of variance (MANOVA) test [37] was based on the three performance measures: true positive volume fraction (TPVF) which indicates the fraction of the total
amount of fluid in the reference standard delineation, false positive volume fraction
(FPVF) [33] which denotes the amount of fluid falsely identified and relative volume
difference ratio (RVDR) which measures the volume difference ratio comparing to
the reference standard volume to show the statistical significance of performance
differences. They are defined as follows,
T PV F
|C T P |
|C td |
(12.9)
F PV F
|C F P |
|U d − C td |
(12.10)
RV D R
||V M |−|V R ||
|V R |
(12.11)
where, U d is assumed to be a binary scene with all voxels in the scene domain set to
have a value 1, and C td is the set of voxels in the true delineation, |·| denotes volume.
|V M | is the segmented volume by method M, and |V R | is the volume of the reference
standard. More details can be seen in [38].
For statistical correlation analysis, we used linear regression analysis [39] and
Bland-Altman plots [40] to evaluate the agreement and relationship between the
automatic and manual segmentations.
For the reproducibility analysis, the retinal specialist was invited to manually
segment the intra- and sub-retinal fluid at the onset of the project and again after
more than 3 months. The manual segmenting of all the slices for one eye required
more than 2 h of expert tracing, so the re-tracing was performed on 5 randomly
selected eyes from the entire data set.
Macula-centered 3D OCT volumes (200 × 200 × 1024 voxels, 6 × 6 × 2 mm
3 ,
voxel size 30 × 30 × 1.95 μm
3 ) were obtained from 15 eyes of 15 patients with
exudative AMD. For the reference standard, a retinal specialist (MDA) manually
355
the sub-graph G SS is labeled as “sink”, i.e. z-S SS (x, y) > d, then a directed arc with a
penalty weight from each node G R (x, y, z) to G SS (x, y, z + d) will be added, as shown
in Fig. 12.6b. For G R and G SI, the same approach is employed.
12.4 Performance Evaluation
12.4.1 Experimental Methods
For the initialization, we compared the performance before and after probability
normalization to show the efficacy of the probability normalization.
For the segmentation, we compared the traditional GC in [13], the traditional
GS method in [29] and the proposed probability constraints GS-GC. A multivariate
analysis of variance (MANOVA) test [37] was based on the three performance measures: true positive volume fraction (TPVF) which indicates the fraction of the total
amount of fluid in the reference standard delineation, false positive volume fraction
(FPVF) [33] which denotes the amount of fluid falsely identified and relative volume
difference ratio (RVDR) which measures the volume difference ratio comparing to
the reference standard volume to show the statistical significance of performance
differences. They are defined as follows,
T PV F
|C T P |
|C td |
(12.9)
F PV F
|C F P |
|U d − C td |
(12.10)
RV D R
||V M |−|V R ||
|V R |
(12.11)
where, U d is assumed to be a binary scene with all voxels in the scene domain set to
have a value 1, and C td is the set of voxels in the true delineation, |·| denotes volume.
|V M | is the segmented volume by method M, and |V R | is the volume of the reference
standard. More details can be seen in [38].
For statistical correlation analysis, we used linear regression analysis [39] and
Bland-Altman plots [40] to evaluate the agreement and relationship between the
automatic and manual segmentations.
For the reproducibility analysis, the retinal specialist was invited to manually
segment the intra- and sub-retinal fluid at the onset of the project and again after
more than 3 months. The manual segmenting of all the slices for one eye required
more than 2 h of expert tracing, so the re-tracing was performed on 5 randomly
selected eyes from the entire data set.
Macula-centered 3D OCT volumes (200 × 200 × 1024 voxels, 6 × 6 × 2 mm
3 ,
voxel size 30 × 30 × 1.95 μm
3 ) were obtained from 15 eyes of 15 patients with
exudative AMD. For the reference standard, a retinal specialist (MDA) manually