Fundus imaging based quantum computing: Grading of severity of vision threatening diabetic retinopathy

Diabetic Retinopathy (DR) is a major complication of diabetes that can result in irreversible vision loss if not detected early. Manual ophthalmologic screening is time-consuming and subject to subjectivity, underscoring the need for accurate and automated diagnostic systems. This study proposes QCNET, a hybrid quantum–classical neural network for multiclass grading of Diabetic Retinopathy from retinal fundus images. The model combines the feature extraction capability of a pre-trained DenseNet201 convolutional network with a four-qubit variational quantum layer that acts as a compact, non-linear feature transformation module within an end-to-end learning framework. The quantum-transformed features are subsequently refined through classical dense layers for final classification. To address class imbalance, advanced data augmentation techniques and random oversampling were used. Interpretability of model is enhanced using Gradient-weighted Class Activation Mapping (Grad-CAM), highlighting clinically relevant retinal regions that influence diagnostic decisions. Experimental evaluation on a publicly available dataset demonstrates that QCNET achieves an overall accuracy of 91.0%, precision of 91.7%, recall of 91.0%, and an F1-score of 90.9%. The Matthews Correlation Coefficient and Cohen’s Kappa values of 0.887 and 0.889 indicate a strong agreement with the ground-truth labels. Class-wise ROC analysis yields AUC values ranging from 0.95 to 1.00, confirming robust discriminative performance across disease stages. These results demonstrate the feasibility and potential of integrating variational quantum circuits within deep learning pipelines for medical image analysis, offering a promising and interpretable decision-support tool for the detection of diabetic retinopathy.