Pathologic myopia causes vision impairment and blindness, and therefore, necessitates a prompt diagnosis. However, there is no standardized definition of pathologic myopia, and its interpretation by 3D optical coherence tomography images is subjective, requiring considerable time and money. Therefore, there is a need for a diagnostic tool that can automatically and quickly diagnose pathologic myopia in patients. This study aimed to develop an algorithm that uses 3D optical coherence tomography volumetric images (C-scan) to automatically diagnose patients with pathologic myopia. The study was conducted using 367 eyes of patients who underwent optical coherence tomography tests at the Ophthalmology Department of Incheon St. Mary’s Hospital and Seoul St. Mary’s Hospital from January 2012 to May 2020. To automatically diagnose pathologic myopia, a deep learning model was developed using 3D optical coherence tomography images. The model was developed using transfer learning based on four pre-trained convolutional neural networks (ResNet18, ResNext50, EfficientNetB0, EfficientNetB4). Grad-CAM was used to visualize features affecting the detection of pathologic myopia. The performance of each model was evaluated and compared based on accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUROC). The model based on EfficientNetB4 showed the best performance (95% accuracy, 93% sensitivity, 96% specificity, and 98% AUROC) in identifying pathologic myopia.

Purpose: In acquiring images of the posterior eye, magnetic resonance imaging (MRI) provides low spatial resolution of the overall shape of the eye while optical coherence tomography (OCT) offers high spatial resolution of the limited range. Through the merger of the two devices, we attempted to acquire detailed anatomy of the posterior eye. Methods: Optical and display distortions in OCT images were corrected using the Listing reduced eye model. The 3.0T orbital MRI images were placed on the three-dimensional coordinate system of the computer-aided design (CAD) program. Employing anterior scleral canal opening, visual axis, and scleral curvature as references, original and corrected OCT images were ported into the CAD application. The radii of curvature of the choroid–scleral interfaces (Rc values) of all original and corrected OCT images were compared to the MRI images. Results: Sixty-five eyes of 33 participants (45.58 ± 19.82 years) with a mean Rc of 12.94 ± 1.24 mm on axial MRI and 13.66 ± 2.81 mm on sagittal MRI were included. The uncorrected horizontal OCT (30.51 ± 9.34 mm) and the uncorrected vertical OCT (34.35 ± 18.09 mm) lengths differed significantly from the MRI Rc values (both P < 0.001). However, the mean Rc values of the corrected horizontal (12.50 ± 1.21 mm) and vertical (13.05 ± 1.98 mm) images did not differ significantly from the Rc values of the corresponding MRI planes (P = 0.065 and P = 0.198, respectively). Conclusions: Features identifiable only on OCT and features only on MRI were successfully integrated into a unitary posterior eye. Translational Relevance: Our CAD-based converging method may establish the collective anatomy of the posterior eye and the neural canal, beyond the range of the OCT.

Purpose: To investigate whether the trajectories of saccadic eye movements (SEMs) significantly differ between glaucoma patients and controls. Methods: SEMs were recorded by video-based infrared oculography in 53 patients with glaucoma and 41 age-matched controls. Participants were asked to bilaterally view 24°-horizontal, 14°-vertical, and 20°-diagonal eccentric Goldmann III-sized stimuli. SEMs were evaluated with respect to the saccadic reaction time (SRT), the mean velocity, amplitude, and two novel measures: departure angle (DA) and arrival angle (AA). These parameters were compared between the groups and the associations of SEM parameters with glaucoma parameters and integrated visual field defects were investigated. Results: Glaucoma patients exhibited increased mean SRT, DA, and AA values compared with controls for 14°-vertical visual targets (P = 0.05, P < 0.01, and P < 0.01, respectively). The SRT, DA, and AA were significantly associated with the mean and pattern standard deviations of perimetry and with the mean RNFL thickness by OCT (all P < 0.001). Glaucoma was associated with the AA (P = 0.05) and both the SRT (P = 0.01) and DA (P = 0.04) were associated with integrated visual field defects. Conclusions: The saccadic trajectories of glaucoma patients depart in an erroneous path and compensate the disparity by deviating the trajectory at arrival. Translational Relevance: The initial deviation that we observed (despite continuous exposure to the stimulus) suggests the disoriented spatial perception of glaucoma patients which may be relevant to difficulties encountered daily.

To develop a smartphone-based pupillometry using deep learning and evaluate its accuracy compared to a commercial pupillometer, the NPi-300. 336 pupillary light reflex (PLR) exams from 158 volunteers were analyzed using deep learning models (UNet, UNet++, DeepLabV3, DeepLabV3+, and Mask R-CNN) with different backbones (ResNet50, Swin Transformer, and ConvNeXt V2). Once the best combination was identified, image data was filtered according to the degree of eyelid opening and image blurriness. The maximum-minimum pupil size difference, constriction velocity (CV), and percentage change in pupil size (CP) were compared between our application and the NPi-300 gold standard. The kernel density estimation and Bhattacharyya Distance were used to develop a scoring method to classify pupil reactivity: SmartPLR. Mask R-CNN (ConvNeXt V2 backbone), which showed a mean intersection over union of 0.9177, segmentation mean average precision (mAP) of 0.8670, and bounding box mAP of 0.8663, was selected for our application. The Pearson correlation values comparing our application to the NPi-300 for pupil size difference, CV, and CP were 0.77, 0.77, and 0.74, respectively. The SmartPLR formula was defined as $$\:{10}^{4}\times\:pred\_diff\times\:pred\_CV\times\:{(pred\_CP)}^{2}$$ , and sluggish pupils were defined as $$\:0.07<\:pred\_CP\le\:0.2$$ and $$\:0.1<\:SmartPLR\le\:5$$ , and urgent pupils were defined as $$\:pred\_CP\le\:0.07$$ and $$\:SmartPLR\le\:0.1$$ . Despite various smartphone applications developed to evaluate the PLR, they rely on additional add-ons or infrared light source. This prevents such applications from being completely commercialized. Our novel smartphone application, built on deep learning and not requiring infrared or additional devices, demonstrated high accuracy compared to the NPi-300.