Limits...
Three-dimensional spectral-domain optical coherence tomography data analysis for glaucoma detection.

Xu J, Ishikawa H, Wollstein G, Bilonick RA, Folio LS, Nadler Z, Kagemann L, Schuman JS - PLoS ONE (2013)

Bottom Line: Features of super pixel map were extracted and used as inputs to machine classifier (LogitBoost adaptive boosting) to automatically identify diseased eyes.The super pixel analysis showed statistically significantly higher AUC than the cpRNFL (0.855 vs. 0.707, respectively, p = 0.031, Jackknife test) when glaucoma suspects were discriminated from healthy, while no significant difference was found when confirmed glaucoma eyes were discriminated from healthy eyes.This new method has the potential to improve early detection of glaucomatous damage.

View Article: PubMed Central - PubMed

Affiliation: University of Pittsburgh Medical Center Eye Center, Eye and Ear Institute, Ophthalmology and Visual Science Research Center, Department of Ophthalmology, University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania, USA.

ABSTRACT

Purpose: To develop a new three-dimensional (3D) spectral-domain optical coherence tomography (SD-OCT) data analysis method using a machine learning technique based on variable-size super pixel segmentation that efficiently utilizes full 3D dataset to improve the discrimination between early glaucomatous and healthy eyes.

Methods: 192 eyes of 96 subjects (44 healthy, 59 glaucoma suspect and 89 glaucomatous eyes) were scanned with SD-OCT. Each SD-OCT cube dataset was first converted into 2D feature map based on retinal nerve fiber layer (RNFL) segmentation and then divided into various number of super pixels. Unlike the conventional super pixel having a fixed number of points, this newly developed variable-size super pixel is defined as a cluster of homogeneous adjacent pixels with variable size, shape and number. Features of super pixel map were extracted and used as inputs to machine classifier (LogitBoost adaptive boosting) to automatically identify diseased eyes. For discriminating performance assessment, area under the curve (AUC) of the receiver operating characteristics of the machine classifier outputs were compared with the conventional circumpapillary RNFL (cpRNFL) thickness measurements.

Results: The super pixel analysis showed statistically significantly higher AUC than the cpRNFL (0.855 vs. 0.707, respectively, p = 0.031, Jackknife test) when glaucoma suspects were discriminated from healthy, while no significant difference was found when confirmed glaucoma eyes were discriminated from healthy eyes.

Conclusions: A novel 3D OCT analysis technique performed at least as well as the cpRNFL in glaucoma discrimination and even better at glaucoma suspect discrimination. This new method has the potential to improve early detection of glaucomatous damage.

Show MeSH

Related in: MedlinePlus

Flowchart of converting a 3D OCT image into a 2D feature map.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC3569462&req=5

pone-0055476-g003: Flowchart of converting a 3D OCT image into a 2D feature map.

Mentions: Each 3D OCT image (200×200×1024 voxels) was converted to a 2D feature map (200×200 pixels) as follows. RNFL thickness and reflectivity along with the blood vessel mapping were extracted from each image (Fig. 3). The normalized RNFL map after applying the bundle path correction were compared with the normative database point by point to obtain a deviation map, with the cut-off value set to mean value minus stardard deviation (SD). Unlike the conventional setting of cut-off value, i.e., mean value - 2SD in most OCT devices, this new setting is more sensitive to identify the case at the border of the normative database, which might be an indicator of the structural changes at the early stage. With this new setting, the RNFL data lower than the bottom 15.9% of the normative database was set with higher probability for the further process, comparing with the bottom 2.3% using the conventional setting. The internal reflectivity of retinal nerve fiber layer has been proved to be useful in glaucoma assessment. [11] The RNFL internal reflectivity, was calculated by taking the average reflectivity within the RNFL along each A-scan, with each voxel’s reflectivity normalized to its A-scan’s saturation before the averaging. Retinal blood vessel generated shadow at retina nerve fiber layer, which was noise in the RNFL thickness computation. Moreover, the vessel patterns varied randomly among subjects. To minimize the vessel effect on the RNFL thickness map, the retinal blood vessels in the 3D dataset were automatically detected using a 3D boosting algorithm [12] and filled out. The RNFL thickness of each pixel located at the blood vessel region was replaced by a value computed from all the non-vessel pixels on the RNFL thickness map using bi-linear interpolation. The final feature map is a function of the RNFL thickness and interal refleciviy after accounting for the blood vessls and the deviation map.


Three-dimensional spectral-domain optical coherence tomography data analysis for glaucoma detection.

Xu J, Ishikawa H, Wollstein G, Bilonick RA, Folio LS, Nadler Z, Kagemann L, Schuman JS - PLoS ONE (2013)

Flowchart of converting a 3D OCT image into a 2D feature map.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC3569462&req=5

pone-0055476-g003: Flowchart of converting a 3D OCT image into a 2D feature map.
Mentions: Each 3D OCT image (200×200×1024 voxels) was converted to a 2D feature map (200×200 pixels) as follows. RNFL thickness and reflectivity along with the blood vessel mapping were extracted from each image (Fig. 3). The normalized RNFL map after applying the bundle path correction were compared with the normative database point by point to obtain a deviation map, with the cut-off value set to mean value minus stardard deviation (SD). Unlike the conventional setting of cut-off value, i.e., mean value - 2SD in most OCT devices, this new setting is more sensitive to identify the case at the border of the normative database, which might be an indicator of the structural changes at the early stage. With this new setting, the RNFL data lower than the bottom 15.9% of the normative database was set with higher probability for the further process, comparing with the bottom 2.3% using the conventional setting. The internal reflectivity of retinal nerve fiber layer has been proved to be useful in glaucoma assessment. [11] The RNFL internal reflectivity, was calculated by taking the average reflectivity within the RNFL along each A-scan, with each voxel’s reflectivity normalized to its A-scan’s saturation before the averaging. Retinal blood vessel generated shadow at retina nerve fiber layer, which was noise in the RNFL thickness computation. Moreover, the vessel patterns varied randomly among subjects. To minimize the vessel effect on the RNFL thickness map, the retinal blood vessels in the 3D dataset were automatically detected using a 3D boosting algorithm [12] and filled out. The RNFL thickness of each pixel located at the blood vessel region was replaced by a value computed from all the non-vessel pixels on the RNFL thickness map using bi-linear interpolation. The final feature map is a function of the RNFL thickness and interal refleciviy after accounting for the blood vessls and the deviation map.

Bottom Line: Features of super pixel map were extracted and used as inputs to machine classifier (LogitBoost adaptive boosting) to automatically identify diseased eyes.The super pixel analysis showed statistically significantly higher AUC than the cpRNFL (0.855 vs. 0.707, respectively, p = 0.031, Jackknife test) when glaucoma suspects were discriminated from healthy, while no significant difference was found when confirmed glaucoma eyes were discriminated from healthy eyes.This new method has the potential to improve early detection of glaucomatous damage.

View Article: PubMed Central - PubMed

Affiliation: University of Pittsburgh Medical Center Eye Center, Eye and Ear Institute, Ophthalmology and Visual Science Research Center, Department of Ophthalmology, University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania, USA.

ABSTRACT

Purpose: To develop a new three-dimensional (3D) spectral-domain optical coherence tomography (SD-OCT) data analysis method using a machine learning technique based on variable-size super pixel segmentation that efficiently utilizes full 3D dataset to improve the discrimination between early glaucomatous and healthy eyes.

Methods: 192 eyes of 96 subjects (44 healthy, 59 glaucoma suspect and 89 glaucomatous eyes) were scanned with SD-OCT. Each SD-OCT cube dataset was first converted into 2D feature map based on retinal nerve fiber layer (RNFL) segmentation and then divided into various number of super pixels. Unlike the conventional super pixel having a fixed number of points, this newly developed variable-size super pixel is defined as a cluster of homogeneous adjacent pixels with variable size, shape and number. Features of super pixel map were extracted and used as inputs to machine classifier (LogitBoost adaptive boosting) to automatically identify diseased eyes. For discriminating performance assessment, area under the curve (AUC) of the receiver operating characteristics of the machine classifier outputs were compared with the conventional circumpapillary RNFL (cpRNFL) thickness measurements.

Results: The super pixel analysis showed statistically significantly higher AUC than the cpRNFL (0.855 vs. 0.707, respectively, p = 0.031, Jackknife test) when glaucoma suspects were discriminated from healthy, while no significant difference was found when confirmed glaucoma eyes were discriminated from healthy eyes.

Conclusions: A novel 3D OCT analysis technique performed at least as well as the cpRNFL in glaucoma discrimination and even better at glaucoma suspect discrimination. This new method has the potential to improve early detection of glaucomatous damage.

Show MeSH
Related in: MedlinePlus