Related Experiment Video
Updated: Aug 6, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
AI for Anterior Segment Disease Using Transfer Learning: Adapting Slit-Lamp-Trained Model for Analysis of Smartphone
Hiroki Maehara1,2, Yuta Ueno2,3, Masahiro Oda4
1Department of Ophthalmology, Tokyo Dental College Ichikawa General Hospital, Chiba, Japan.
Translational Vision Science & Technology
|July 17, 2026
Summary
We developed a Phone-tuned artificial intelligence (AI) model using smartphone images and transfer learning. This AI significantly improves anterior segment corneal disease classification accuracy, aiding remote diagnostics.
Area of Science:
- Ophthalmology
- Artificial Intelligence
- Medical Imaging
Background:
- Existing artificial intelligence (AI) models for anterior segment corneal diseases are trained solely on slit-lamp images.
- This limits their applicability in real-world scenarios where smartphone imaging is more accessible.
Purpose of the Study:
- To adapt an existing AI model for anterior segment corneal disease classification to effectively utilize smartphone images.
- To evaluate the performance of an AI model fine-tuned with smartphone images via transfer learning.
Main Methods:
- Transfer learning was applied to two AI models (YOLOv5 and YOLOX) using 2530 smartphone-captured corneal images.
- The accuracy of the original and Phone-tuned AI models was evaluated and compared.
Main Results:
- Phone-tuned AI (YOLOv5) achieved a statistically significant improvement in average accuracy (93.5%) compared to the original AI (67.0%).
- Phone-tuned AI (YOLOX) showed a slight accuracy increase (84.2% vs. 78.4%), though not statistically significant.
- The Phone-tuned AI (YOLOv5) demonstrated high accuracy across different urgency levels for corneal diseases.
Conclusions:
- Transfer learning with smartphone images significantly enhances AI model performance for corneal disease classification, particularly with the YOLOv5 architecture.
- Phone-tuned AI shows promise for improving diagnosis and triage, especially in underserved areas lacking ophthalmologists.