The Role of Quantum-enhanced Support Vector Machine using Multiparametric MRI Parameters in Differentiating Medulloblastoma from Ependymoma

Brain tumors are the leading cause of death from solid tumors in the childhood. The most common pediatric brain cancers are posterior fossa tumors, with Medulloblastoma (MB) and Ependymoma (EP) being the most common. Although the treatment and prognosis of MB and EP are different, visual characteristics of these tumors are often overlapping, sometimes making the diagnostic process difficult. Therefore, the aim of the current study is twofold: (i) to conduct a comparative analysis with support vector machine (SVM) algorithm and quantum-enhanced support vector machine (QSVM) using multiparametric (mp) MRI of tissue characteristics of whole lesion outlining its diffusivity, cellularity, and vascularity in differentiating MB from EP; and (ii) to explore the effect of noise data on the QSVM. The screening MR examinations of 49 children; MB (n=40) and EP (n=9) were included in the analysis. We used semi quantitative MR imaging features, which give information about the vascularity characteristics of lesion. We utilized precomputed quantum kernel and fed it into classical SVM to classify these tumors. The data gets mapped to high dimensional Hilbert space with PauliFeatureMap. In addition, we further compared the classification performance of QSVM with classical SVM. In the next part of the study, Synthetic Minority Oversampling Technique (SMOTE) method was used for balancing the imbalanced data points of MB and EP. Here, we investigated the performance of QSVM on noisy data. QSVM gave the superior performance with 90% test score value. Classical SVM compared to QSVM, gave the equal performance with 90% for linear, polynomial, radial basis function and sigmoid kernels and lower performance with 80% for poly kernel. In the case of noisy data, the performance of QSVM fell to 50%. Our study confirms that the QSVM method could be used for posterior fossa brain tumor classification problems with higher performance.