Quantum Machine Learning for Biomedical Classification Problems: A Feasibility Study on Real Quantum Hardware

Purpose: Recent advances in quantum computing offer opportunities to explore alternative methods for the solution of classification problems commonly found in biomedical research. This study investigates the feasibility of using quantum kernel-based Support Vector Machines (QSVMs) to classify Autism Spectrum Disorder (ASD) using metabolomic measurements on real quantum hardware. This work evaluates the capabilities of current quantum computers for biomedical classification and establishes practical baselines for future studies.

Methods: A quantum classification pipeline was developed using a variety of angle encoding schemes. An exhaustive search was performed to identify an optimal subset of four metabolomic features via simulation. These features were used to benchmark multiple encoding strategies via simulation, followed by validation on IBM Quantum hardware. A baseline using Support Vector Machine (SVM) with the same features was established for comparison.

Results: The best-performing QSVM achieved an average classification accuracy of 0.9434 on real quantum hardware, which is comparable to the accuracy using classical SVM of 0.9371 on the same feature set. These results highlight the potential of quantum kernels to capture meaningful feature interactions in biomedical data, despite the levels of noise and overhead of quantum computing.

Conclusion: This study demonstrates that quantum kernel SVMs can achieve classification performance comparable to classical methods on biomedical data. However, current limitations in quantum hardware, such as qubit communication overhead and noise, pose challenges for practical deployment. Continued improvements in hardware acceleration and error correction are needed to realize the potential of quantum machine learning for biomedical classification tasks.