Clinical data classification with noisy intermediate scale quantum computers.

Area of Science:

• Quantum Computing
• Machine Learning
• Computational Biology

Background:

• Quantum machine learning (QML) is rapidly advancing, showing promise for near-term quantum computers.
• The application of QML to real-world clinical datasets remains an active area of research.

Purpose of the Study:

• To investigate the feasibility of using QML algorithms for data classification on real clinical datasets.
• To propose and evaluate two novel QML algorithms: a quantum distance classifier (qDS) and a simplified quantum-kernel support vector machine (sqKSVM).

Main Methods:

• Implemented qDS and sqKSVM algorithms on the 15-qubit IBMQ Melbourne quantum computer.
• Utilized a linear time quantum data encoding technique for embedding classical data into quantum states.
• Compared QML performance against classical algorithms and prior QML methods on three open-access clinical datasets.

Main Results:

• The qDS algorithm outperformed kernel-based methods on datasets with small sample and feature counts.
• Quantum kernel approaches demonstrated superior performance on datasets with high sample and feature counts.
• The linear time quantum data encoding technique improved predictive performance by up to 2% in area under the receiver operator characteristics curve.

Conclusions:

• QML algorithms show potential for clinical data analysis, with performance varying based on dataset characteristics.
• The proposed quantum data encoding method enhances QML predictive accuracy, making it suitable for Noisy Intermediate Scale Quantum (NISQ) computers.