Quantum SVM-driven framework for accurate brain stroke classification

Area of Science:

• Medical Imaging and Diagnostics
• Quantum Computing Applications
• Computational Neuroscience

Background:

• Brain stroke is a critical cerebrovascular disorder causing irreversible neuronal damage if not diagnosed promptly.
• Early and accurate stroke subtype classification is crucial for reducing mortality and improving patient recovery.
• Traditional diagnostic methods like MRI and CT scans involve time-consuming manual interpretation prone to subjective variability.

Purpose of the Study:

• To develop and evaluate an automated brain stroke classification framework using Quantum Support Vector Machines (QSVM).
• To integrate classical feature extraction with quantum kernel methods for enhanced classification performance.
• To compare the QSVM framework's efficacy against established classical machine learning algorithms.

Main Methods:

• A unified classical feature extraction pipeline was employed, including textural, morphological, frequency-domain, and statistical descriptors.
• Classical features were encoded into quantum states using a six-qubit circuit and a ZZFeatureMap-based quantum kernel.
• The quantum kernel mapped features into a higher-dimensional Hilbert space to improve nonlinear separability.
• Stratified 5-fold cross-validation was used for experimental evaluation on a public Kaggle MRI dataset using a classical quantum simulator.

Main Results:

• The QSVM framework achieved high classification performance: 96.8% accuracy, 96.2% precision, 97.1% recall, and a 96.6% F1-score.
• An AUC-ROC of 0.982 was recorded, indicating strong discriminative ability.
• The QSVM outperformed optimized classical baselines (Random Forest, KNN, SVM) on the same feature sets.

Conclusions:

• Quantum-inspired kernel methods, like the proposed QSVM, show potential for improving brain stroke classification performance.
• The framework demonstrates significant simulator-based performance gains, suggesting a promising direction for automated diagnostics.
• Further validation on larger, multicenter datasets and real quantum hardware is warranted to confirm hardware-level quantum advantage.