Deep Learning Model for Static Ocular Torsion Detection Using Synthetically Generated Fundus Images.

Purpose: The objective of the study is to develop deep learning models using synthetic fundus images to assess the direction (intorsion versus extorsion) and amount (physiologic versus pathologic) of static ocular torsion. Static ocular torsion assessment is an important clinical tool for classifying vertical ocular misalignment; however, current methods are time-intensive with steep learning curves for frontline providers. Methods: We used a dataset (n = 276) of right eye fundus images. The disc-foveal angle was calculated using ImageJ to generate synthetic images via image rotation. Using synthetic datasets (n = 12,740 images per model) and transfer learning (the reuse of a pretrained deep learning model on a new task), we developed a binary classifier (intorsion versus extorsion) and a multiclass classifier (physiologic versus pathologic intorsion and extorsion). Model performance was evaluated on unseen synthetic and nonsynthetic data. Results: On the synthetic dataset, the binary classifier had an accuracy and area under the receiver operating characteristic curve (AUROC) of 0.92 and 0.98, respectively, whereas the multiclass classifier had an accuracy and AUROC of 0.77 and 0.94, respectively. The binary classifier generalized well on the nonsynthetic data (accuracy = 0.94; AUROC = 1.00). Conclusions: The direction of static ocular torsion can be detected from synthetic fundus images using deep learning methods, which is key to differentiate between vestibular misalignment (skew deviation) and ocular muscle misalignment (superior oblique palsies). Translational Relevance: Given the robust performance of our models on real fundus images, similar strategies can be adopted for deep learning research in rare neuro-ophthalmologic diseases with limited datasets.

Caregiver Preferences for Three-Dimensional Printed or Augmented Reality Craniosynostosis Skull Models: A Cross-Sectional Survey.

BACKGROUND: Recent advances in three-dimensional (3D) printing and augmented reality (AR) have expanded anatomical modeling possibilities for caregiver craniosynostosis education. The purpose of this study is to characterize caregiver preferences regarding these visual models and determine the impact of these models on caregiver understanding of craniosynostosis. METHODS: The authors constructed 3D-printed and AR craniosynostosis models, which were randomly presented in a cross-sectional survey. Caregivers rated each model's utility in learning about craniosynostosis, learning about skull anatomy, viewing an abnormal head shape, easing anxiety, and increasing trust in the surgeon in comparison to a two-dimensional (2D) diagram. Furthermore, caregivers were asked to identify the fused suture on each model and indicate their preference for generic versus patient-specific models. RESULTS: A total of 412 craniosynostosis caregivers completed the survey (mean age 33 years, 56% Caucasian, 51% male). Caregivers preferred interactive, patient-specific 3D-printed or AR models over 2D diagrams (mean score difference 3D-printed to 2D: 0.16, P < 0.05; mean score difference AR to 2D: 0.17, P < 0.01) for learning about craniosynostosis, with no significant difference in preference between 3D-printed and AR models. Caregiver detection accuracy of the fused suture on the sagittal model was 19% higher with the 3D-printed model than with the AR model (P < 0.05) and 17% higher with the 3D-printed model than with the 2D diagram (P < 0.05). CONCLUSIONS: Our findings indicate that craniosynostosis caregivers prefer 3D-printed or AR models over 2D diagrams in learning about craniosynostosis. Future craniosynostosis skull models with increased user interactivity and patient-specific components can better suit caregiver preferences.