FQPDR: Federated Quantum Neural Network Detects Diabetic Retinopathy Privately
Quantum + federation detects tiny microaneurysm dots with limited data and fewer parameters.
Diabetic retinopathy (DR) is a leading cause of blindness, and early detection hinges on spotting tiny, low-contrast microaneurysm dots. A new paper on arXiv (arXiv:2605.08324) introduces FQPDR, a federated quantum neural network that tackles this challenge while preserving patient privacy. Developed by Debashis De, Mahua Nandy Pal, and Dipankar Hazra, the model combines federated learning—which shares only model parameters, not raw data—with quantum neural networks to achieve high detection accuracy with exceptionally limited training samples and few learnable parameters.
The team validated FQPDR on the E-ophtha and Retina MNIST datasets, then cross-evaluated on Kaggle images. Results show that this lightweight quantum approach outperforms both traditional non-federated and federated methods in detecting early-stage DR markers. The system's efficiency stems from quantum computing's ability to handle high-dimensional feature spaces with fewer resources, while federation ensures sensitive medical images never leave local institutions. This work, published in *Evolutionary Intelligence* (2024), represents a significant step toward deployable, privacy-compliant AI in ophthalmology.
- FQPDR uses a quantum neural network with federated learning to detect microaneurysm dots from diabetic retinopathy.
- Trained on E-ophtha and Retina MNIST datasets with limited samples and few learnable parameters.
- Cross-evaluation on Kaggle images shows robust performance exceeding existing non-FL and FL methods.
Why It Matters
Enables privacy-preserving, early DR screening with quantum efficiency, reducing blindness risk across institutions.