Error mitigation enables PET radiomic cancer characterization on quantum computers.

Area of Science:

• Quantum computing applications in medicine
• Radiomics and medical imaging analysis
• Machine learning for clinical endpoint prediction

Background:

• Cancer diagnosis relies on biopsy, which has limitations in characterizing tumor heterogeneity.
• Positron emission tomography (PET)-driven radiomics shows potential for predicting clinical outcomes.
• Quantum machine learning (QML) is explored for enhanced predictive capabilities in cancer patients.

Purpose of the Study:

• To evaluate the added value of quantum machine learning (QML) in predicting clinical endpoints using PET radiomics data.
• To compare QML performance against classical machine learning (CML) on both simulated and real quantum computers.
• To investigate the impact of error mitigation techniques on QML predictive accuracy.

Main Methods:

• Utilized publicly available PET radiomics datasets for glioma, prostate, and lung cancer.
• Applied redundancy reduction and feature selection, creating 18 dataset variants.
• Trained and tested five CML and their QML counterparts in simulators, and selected QML models on the IonQ Aria quantum computer with error mitigation.

Main Results:

• QML generally outperformed CML in simulator environments, especially with 16 features (70% vs. 69% BACC).
• Error mitigation significantly improved QML performance on the IonQ device, increasing test BACC from 69.94% to 75.66%.
• Quantum advantage was observed in simulator environments and on real quantum hardware when utilizing QML with error mitigation.

Conclusions:

• Quantum advantage is achievable in real quantum computers for predicting clinical endpoints in PET cancer cohorts, particularly with error mitigation.
• QML demonstrates potential for improved predictive accuracy in cancer patient stratification using radiomic data.
• This study highlights the feasibility and benefits of applying quantum computing to medical diagnostics and prognostics.