Abstract
Purpose: :
In this work, the ability of a visual pathway analysis to differentiate between glaucoma and healthy subjects is examined. A system to detect glaucoma based on histogram features of the diffusion tensor derived indices in the optic radiation is proposed and validated.
Methods: :
Diffusion tensor Imaging (DTI) brain scans of 59 subjects were acquired. The subjects were categorized into two age matched groups: 19 controls (age 63.3 ± 8.1 years) and 40 subjects with primary open angle glaucoma (age 64.7 ± 7.5 years).The optic radiation was initially identified using the authors' previously published automated segmentation algorithm. A region of interest (ROI) representing the optic radiation in a single slice that includes the termination of the optic tract in the lateral geniculate nucleus was selected. The segmented optic radiation was manually corrected by two DTI experts. The apparent diffusion coefficient (ADC), axial and radial (RD) diffusivities, as well as fractional anisotropy were derived from the diffusion tensor and used to characterize the selected ROI. The histograms of the DTI-derived parameters were calculated. The following features were extracted from the histograms: Mean, variance, skewness, kurtosis, energy, and entropy. A support vector machine classifier was used to rank the 24 histogram features based on their individual abilities in discriminating glaucoma patients from controls. The three highest ranked features in addition to the means of ADC and RD were used as features for a logistic regression classifier. The performance of the classification was evaluated using a 5-fold cross validation setup.
Results: :
The proposed system achieved a classification accuracy of 81.4%. The area under the receiver operating characteristic (ROC) curve was found to be 0.84 with a specificity of 63.2% and a sensitivity of 90% for glaucoma detection.
Conclusions: :
The presented analysis showed that the DTI-derived indices (characterizing different aspects of the fiber structure) provide competitive glaucoma classification rates compared to the rates based on eye imaging modalities. This work presents a new perspective in identifying glaucoma using visual pathway analysis. Furthermore, it is complementary to conventional eye examinations and could lead to improvements in glaucoma detection and diagnosis.
Keywords: image processing • visual impairment: neuro-ophthalmological disease • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound)