Open-Source Automated Segmentation of Neuronal Structures in Corneal Confocal Microscopy Images of the Subbasal Nerve Plexus With Accuracy on Par With Human Segmentation.

PURPOSE: The aim of this study was to perform automated segmentation of corneal nerves and other structures in corneal confocal microscopy (CCM) images of the subbasal nerve plexus (SNP) in eyes with ocular surface diseases (OSDs). METHODS: A deep learning-based 2-stage algorithm was designed to perform segmentation of SNP features. In the first stage, to address applanation artifacts, a generative adversarial network-enabled deep network was constructed to identify 3 neighboring corneal layers on each CCM image: epithelium, SNP, and stroma. This network was trained/validated on 470 images of each layer from 73 individuals. The segmented SNP regions were further classified in the second stage by another deep network as follows: background, nerve, neuroma, and immune cells. Twenty-one-fold cross-validation was used to assess the performance of the overall algorithm on a separate data set of 207 manually segmented SNP images from 43 patients with OSD. RESULTS: For the background, nerve, neuroma, and immune cell classes, the Dice similarity coefficients of the proposed automatic method were 0.992, 0.814, 0.748, and 0.736, respectively. The performance metrics for automatic segmentations were statistically better or equal as compared to human segmentation. In addition, the resulting clinical metrics had good to excellent intraclass correlation coefficients between automatic and human segmentations. CONCLUSIONS: The proposed automatic method can reliably segment potential CCM biomarkers of OSD onset and progression with accuracy on par with human gradings in real clinical data, which frequently exhibited image acquisition artifacts. To facilitate future studies on OSD, we made our data set and algorithms freely available online as an open-source software package.

Test of a Retinal Nerve Fiber Bundle Trajectory Model Using Eyes With Glaucomatous Optic Neuropathy.

Purpose: To test a model of retinal nerve fiber bundle trajectories that predicts the arcuate-shaped patterns seen on optical coherence tomography (OCT) retinal nerve fiber layer (RNFL) probability/deviation maps (p-maps) in glaucomatous eyes. Methods: Thirty-one glaucomatous eyes from a database of 250 eyes had clear arcuate-shaped patterns on RNFL p-maps derived from an OCT cube scan. The borders of the arcuate patterns were extracted from the RNFL p-maps. Next, the trajectories from an arcuate model were compared against these borders via a normalized root-mean-square difference analysis. The model's parameter, ß, was varied, and the best-fitting, initial clock-hour position of the trajectory to the border was found for each ß. Finally, the regions, as determined by the arcuate border's best-fit, initial clock-hour positions, were compared against the abnormal regions on the circumpapillary retinal nerve fiber layer (cpRNFL) profile. Results: The arcuate model's mean ßSup and ßInf parameters minimized large differences between the trajectories and the arcuate borders on the RNFL p-maps. Furthermore, on average, 68% of the cpRNFL regions defined by the arcuate border's best-fit, initial clock-hour positions were abnormal (i.e., below the ≤5% threshold). Conclusions: The arcuate model performed well in predicting the borders of arcuate patterns seen on RNFL p-maps. It also predicted the associated abnormal regions of the cpRNFL thickness plots. Translational Relevance: This model should prove useful in helping clinicians understand topographical comparisons among different OCT representations and should improve structure-structure, as well as structure-function agreement analyses.