# Automated eye disease classification method from anterior eye image using anatomical structure focused image classification technique

**Authors:** Masahiro Oda, Takefumi Yamaguchi, Hideki Fukuoka, Yuta Ueno, Kensaku Mori  
**arXiv:** [2005.01433](https://arxiv.org/abs/2005.01433)  
**Date:** 2020-05-04  
**Summary page:** https://www.summarizepaper.com/en/paper/2005.01433v1/

## Key points

- Automated classification method for distinguishing between infective and non-infective diseases from anterior eye images
- Differentiation between these two types of diseases is crucial for determining appropriate treatment plan
- Currently, ophthalmologists rely on empirical observations for distinction
- Quantitative classification system with computer assistance is necessary for accurate diagnosis and treatment
- Challenges in anterior eye images due to variations in eye position and illumination brightness
- Proposed anatomical structure focused image classification technique targeting the cornea
- Leveraging YOLOv3 object detection method to detect opacified areas in corneas associated with diseases
- Achieved an impressive accuracy rate of 88.3% in correctly classifying the images
- Potential to assist ophthalmologists in accurately diagnosing and treating anterior eye diseases

## Summary

This paper introduces an automated classification method for distinguishing between infective and non-infective diseases from anterior eye images. The differentiation between these two types of diseases is crucial in determining the appropriate treatment plan. Currently, ophthalmologists rely on empirical observations to make this distinction; however, a quantitative classification system with computer assistance is necessary for accurate diagnosis and treatment. Anterior eye images present challenges due to variations in eye position and illumination brightness. To overcome these difficulties, the authors propose an anatomical structure focused image classification technique that targets the cornea. They leverage the YOLOv3 object detection method to detect opacified areas in the corneas associated with infective and non-infective diseases. In their experiments using anterior eye images, the proposed method achieved an impressive accuracy rate of 88.3% in correctly classifying the images. This automated approach has significant potential to assist ophthalmologists in accurately diagnosing and treating anterior eye diseases. The authors of this paper are Masahiro Oda, Takefumi Yamaguchi, Hideki Fukuoka, Yuta Ueno, and Kensaku Mori. The paper was accepted as a poster presentation at SPIE Medical Imaging 2020 and published in Proceedings of SPIE Medical Imaging 2020: Computer-Aided Diagnosis (Vol.11314, 1131446).