Akademik Veri Yönetim Sistemi
Neural Computing and Applications, 2025 (SCI-Expanded)
Early detection and accurate classification of brain tumors using MRI scans are crucial for effective diagnosis and treatment planning. However, with the growing patient population and the increasing volume of MRI data, as well as limitations like noise in image data and poor resolution, accurate and rapid diagnosis becomes challenging. To address these issues, AI systems are needed to support radiologists by offering a second opinion. Recent advancements in deep learning (DL) have significantly improved MRI-based brain tumor diagnosis. Despite these improvements, challenges such as the need for higher computational power, difficulty processing large and high-resolution datasets, and limitations of classical vector space. However, quantum computing and quantum computing-based AI methods, by leveraging properties such as superposition and entanglement, have the potential to process data in parallel, handle higher-dimensional data more efficiently, and solve certain problems that classical methods struggle with, more quickly and efficiently. In this study, we proposed four different hybrid quantum–classical integrated neural network (HQCINN) models featuring various multilayer parameterized quantum circuit architectures, which we refer to as “shallow and deep circuits,” designed based on properties such as “entanglement capability, circuit loss, and the number of trainable parameters.” These models aim to distinguish between glioma, meningioma, pituitary and non-tumor classes. The performance of these models was compared to classical DL models, revealing that quantum models provide higher accuracy and lower loss values with fewer parameters. Additionally, when the HQCINN model with the best performance was applied to a brain tumor dataset consisting of CT images, it demonstrated consistent performance across different patient data distributions and imaging modalities, thereby showing strong generalization capability. These results suggest that HQCINN approaches could provide significant advantages in medical imaging tasks, particularly in complex datasets like brain tumor classification.