Welcome to our project on cataract detection from fundus images!
Cataract is an eye disorder characterized by the clouding of the lens, leading to impaired vision. It stands as the leading cause of visual impairment globally, with nearly 40% of blind individuals affected by cataracts. Traditional approaches to automatically detect this condition using Artificial Intelligence have encountered several challenges, including low precision rates and model overfitting due to limited data.
Our project addresses these challenges by proposing a robust model that leverages the power of three prominent classifiers: InceptionV3, VGG19, and InceptionResNetV2. We utilize a stacking technique to combine their predictions into a meta-model based on a neural network, aiming for optimal performance.
The results from our model demonstrate significant performance improvements:
- Accuracy: 98.31%
- Specificity: 100%
- Sensitivity: 97.44%
- Precision: 100%
- F1 Score: 98.70%
These metrics highlight the model's effectiveness in accurately distinguishing between cataract and normal cases.
- Advanced Stacking Approach: Combines predictions from InceptionV3, VGG19, and InceptionResNetV2.
- High Precision and Sensitivity: Ensures accurate detection with minimal false positives and false negatives.
- Robust Neural Network Meta-Model: Enhances overall model performance and reliability.
Machine Learning, Stacking, Cataract, Convolutional Neural Networks, Ophthalmology
We are excited to share our work with the community and look forward to collaborative efforts to further advance the field of automated cataract detection. Explore our repository to learn more about the implementation details, datasets used, and how to contribute to this project.