Benchmarking MedMNIST dataset on real quantum hardware.

Area of Science:

• Quantum Computing
• Machine Learning
• Medical Imaging

Background:

• Quantum machine learning (QML) offers potential for complex classification tasks.
• Medical image analysis requires robust computational methods.

Purpose of the Study:

• To benchmark QML models on real quantum hardware for medical image classification.
• To evaluate the feasibility and performance of quantum-only models on the MedMNIST dataset.
• To explore advanced quantum computing techniques for practical healthcare applications.

Main Methods:

• Preprocessing medical images to reduce spatial dimensions.
• Generating hardware-efficient, noise-resilient quantum circuits.
• Optimizing and training QML models classically, followed by inference on IBM quantum hardware.
• Implementing error suppression and mitigation techniques: dynamical decoupling (DD), gate twirling (Twir), and matrix-free measurement mitigation (M3).

Main Results:

• Demonstrated the performance of QML models on a 127-qubit IBM quantum processor.
• Showcased the effectiveness of error mitigation techniques in improving classification accuracy.
• Established a benchmark for QML in medical imaging.

Conclusions:

• Quantum computing shows promise for medical imaging tasks.
• QML models, with appropriate error handling, are feasible for practical healthcare applications.
• This work sets a foundation for future advancements in quantum-assisted medical diagnostics.