Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Effect of Signal-Strength Filtering on 3D Convolutional Neural Network-Based Visual Field Estimation From Macular

Makoto Koyama1, Hidenori Takahashi2,3, Satoru Inoda3

  • 1Minamikoyasu Eye Clinic, Kimitsu, Japan.

Translational Vision Science & Technology
|July 13, 2026
PubMed
Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same authorSame journal

Multicenter OCT-Based Visual Field Representations via Segmentation-Free 3D CNNs: Forecasting, Longitudinal Variability, and Progression Detection.

Translational vision science & technology·2026
Same author

Aqueous humour cytokine profiles in retinal vein occlusion: a comparative study of acute, chronic recurrent, and control eyes.

Graefe's archive for clinical and experimental ophthalmology = Albrecht von Graefes Archiv fur klinische und experimentelle Ophthalmologie·2026
Same author

Colorectal liver metastasis with bile duct tumor thrombus discovered 15 years post primary tumor resection: a case report and literature review.

International cancer conference journal·2023
Same author

Change Impact of Body Composition During Neoadjuvant Chemoradiotherapy in Patients with Resectable and Borderline Resectable Pancreatic Ductal Adenocarcinoma Undergoing Pancreatectomy.

Annals of surgical oncology·2022
Same author

ASO Author Reflections: Neoadjuvant Chemoradiotherapy for Patients with Resectable and Borderline Resectable Pancreatic Ductal Adenocarcinoma Undergoing Pancreatectomy Changes Body Composition Influencing Long-term Outcomes.

Annals of surgical oncology·2022
Same author

The utility of bile juice culture analysis for the management of postoperative infection after pancreaticoduodenectomy.

Surgery·2022

Including all signal strengths in deep learning models improves visual field estimation from optical coherence tomography scans, especially for lower-quality data. This enhances model robustness without penalizing performance on higher-quality scans.

Area of Science:

  • Ophthalmology
  • Artificial Intelligence
  • Medical Imaging

Background:

  • Deep learning models are increasingly used for visual field (VF) estimation from optical coherence tomography (OCT) scans.
  • The quality of OCT scans, indicated by signal strength indices (SSIs), can influence model performance.
  • Excluding low-quality scans might improve model accuracy but could limit generalizability.

Purpose of the Study:

  • To evaluate whether excluding low-signal-strength OCT scans from training improves deep learning-based VF estimation.
  • To determine the impact of including all OCT signal strengths on model performance across different quality levels.

Main Methods:

  • Retrospective analysis of 79,803 paired OCT and VF examinations from 8511 patients.
  • A 3D convolutional neural network (3DCNN) was trained on all scans versus scans with SSIs ≥ 7.

Related Experiment Videos

  • Models were compared using pointwise mean absolute error (MAE) and absolute mean deviation (MD) error on held-out samples stratified by SSI.

    • Main Results:

    • The model trained on all SSIs showed significantly lower errors in the low-SSI subgroup (SSI < 7) for all endpoints.
    • No significant performance differences were observed between models in the high-SSI subgroup (SSI ≥ 7).
    • Error increased as SSI decreased, without a clear threshold at SSI = 7.

    Conclusions:

    • Training deep learning models with OCT scans across the full SSI range is associated with lower error in low-SSI held-out scans.
    • This approach improves robustness to lower-quality inputs without penalizing performance on higher-quality scans.
    • Including lower-signal-strength OCT scans enhances the robustness of segmentation-free OCT-based VF estimation frameworks.