Get started
Work

Fast and Large-Scale Brain Hemorrhage Detection Using RKNODE U-Net and Enhanced Blending RKNODE M-Net with Cloud Integration

Year: 2025
Type: article
Abstract: Brain hemorrhage, a critical kind of stroke resulting from ruptured blood vessels, necessitates prompt identification and intervention to mitigate death rates. This research presents a rapid and scalable method for bleeding detection utilizing the RSNA brain hemorrhaging dataset, integrating sophisticated deep learning techniques with a cloud-based platform for effective training, storage, and global accessibility. A hybrid approach employing ResNet50, DenseNet121, and VGG16 is implemented for feature extraction, while a novel quantum-behaved particle swarm optimizing technique utilizing differential equations is introduced for feature selection, enabling efficient exploration, reduced density, and stable convergence. The chosen characteristics are integrated and classified utilizing a fourth-order Runge-Kutta Neural ODE meta-network, thereby augmenting classification resilience via adaptive depth modelling. A U-Net architecture augmented with a Runge–Kutta ODE block in the bottleneck is employed for RSNA CT image segmentation to accurately localize hemorrhagic regions, enabling segmentation-guided feature learning that enhances downstream classification performance despite the absence of pixel-level annotations in the RSNA dataset. The segmented regions further allow estimation of hemorrhage size and localization. The results of experiments on benchmark datasets indicate enhanced classification and segmentation accuracy, less redundancy, increased prediction speed and improved efficiency compared to traditional methods, underscoring the framework's possibility of real-time large-scale clinical application.
DOI: 10.22399/ijcesen.4820
Language: English
Open Access status: diamond
OpenAlex ID: W7127052573
FWCI:
Cites: 30
Cited by:
Related to: 10