Related Experiment Video
Updated: Jun 25, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
CSDNet: A Novel Deep Learning Framework for Improved Cataract State Detection
Lahari P L1, Ramesh Vaddi1, Mahmoud O Elish2,3
1Department of Electronics and Communication Engineering, SRM University AP, Andhra Pradesh, India.
Diagnostics (Basel, Switzerland)
|May 24, 2024
Summary
A new deep learning model, CSDNet, efficiently detects cataract states with high accuracy. This lightweight framework is ideal for real-time applications and devices with limited resources.
Area of Science:
- Ophthalmology
- Computer Science
- Artificial Intelligence
Background:
- Cataracts are a leading cause of vision impairment and blindness globally.
- Current diagnostic methods face challenges in speed and accessibility.
- Deep learning offers potential for improved automated detection.
Purpose of the Study:
- To develop a lightweight and adaptable deep learning framework for cataract state detection.
- To reduce computational costs and enable real-time or near-real-time inference.
- To improve the accuracy and efficiency of cataract diagnosis.
Main Methods:
- Utilized the Ocular Disease Intelligent Recognition (ODIR) database for training and testing.
- Developed the Cataract States Detection Network (CSDNet) with smaller kernels and fewer parameters.
- Compared CSDNet's performance against established models like VGG19, ResNet50, and EfficientNet B0.
Main Results:
- Achieved 97.24% binary classification accuracy (normal vs. cataract).
- Attained 98.17% accuracy in detecting four cataract states.
- CSDNet is a lightweight 17 MB model with 175,617 trainable parameters and a 212 ms runtime.
Conclusions:
- CSDNet provides a highly accurate and efficient solution for cataract detection.
- The model's lightweight design makes it suitable for resource-constrained environments.
- CSDNet is well-suited for real-time ophthalmic screening and diagnosis.
