Inspiration
Glaucoma is a leading cause of irreversible blindness and often progresses silently without early symptoms. Late diagnosis can result in permanent vision loss, while access to ophthalmologists is limited in many rural and underserved areas.
We were inspired to develop GlaucoScan XAI as an accessible AI-based screening solution that can analyze retinal fundus images and provide an understandable explanation of its prediction.
What it does
GlaucoScan XAI is an offline AI-powered glaucoma screening system. It takes a retinal fundus image as input, preprocesses the image, and uses a Convolutional Neural Network (CNN) to classify glaucoma risk.
The system also generates an Explainable AI (XAI) heatmap that highlights the retinal regions that influenced the model's prediction. A screening report with the prediction and visual explanation is then provided to the user.
How we built it
The system was developed using Python with TensorFlow/Keras for CNN-based classification, OpenCV for image preprocessing, and Gradio for the user interface.
The workflow consists of fundus image input, preprocessing, CNN-based feature extraction and prediction, XAI heatmap generation, and screening report generation. The system is designed to operate offline, making it suitable for remote and field-based screening environments.
Challenges we faced
One of the main challenges was designing an AI system that was not only capable of image classification but also provided an understandable explanation of its prediction. Integrating explainable visualization with the screening workflow was important for improving transparency and interpretability.
Another challenge was designing the system for offline operation while keeping the interface simple and practical for screening use.
What we learned
Through this project, we learned how CNNs can be applied to medical image analysis and how Explainable AI can make AI-based predictions more transparent. We also gained experience in image preprocessing, model development, interface design, and building an offline healthcare screening workflow.
Future scope
Future improvements include mobile deployment using TensorFlow Lite, training with larger datasets, stage-wise glaucoma detection, and integration with clinical systems.
GlaucoScan XAI is designed as a screening-support tool and is not intended to replace professional medical diagnosis.
Log in or sign up for Devpost to join the conversation.