QuantumNet: An enhanced diabetic retinopathy detection model using classical deep learning-quantum transfer learning

Area of Science:

• Ophthalmology
• Medical Imaging
• Quantum Computing

Background:

• Diabetic Retinopathy (DR) is a diabetes-related eye condition that damages retinal blood vessels, potentially causing vision loss.
• Early and accurate DR diagnosis is crucial but challenging due to subtle and varied symptoms.
• Classical deep learning (DL) models have limitations in resource efficiency and accuracy for DR detection.

Purpose of the Study:

• To introduce QuantumNet, a novel hybrid model combining classical DL with quantum transfer learning for enhanced DR detection.
• To evaluate the performance of QuantumNet against established classical DL models.
• To demonstrate the potential of quantum computing in improving medical imaging diagnostics.

Main Methods:

• Classical DL models (CNN, ResNet50, MobileNetV2) were evaluated on the APTOS 2019 dataset.
• The best-performing classical model was integrated with a variational quantum classifier for QuantumNet.
• Quantum transfer learning was employed, with validation using statistical metrics and Google Cirq.

Main Results:

• QuantumNet achieved an accuracy of 94.11% in DR detection.
• This represents an 11.93 percentage point improvement over existing classical DL models.
• The hybrid model demonstrated high accuracy and resource efficiency.

Conclusions:

• QuantumNet offers a transformative solution for accurate and efficient diabetic retinopathy detection.
• The study highlights the significant potential of hybrid quantum-classical approaches in medical imaging.
• QuantumNet paves the way for broader applications of quantum computing in healthcare diagnostics.