3B Rubik Evri simlerle Medikal G r nt B l tleme

Accurate segmentation of volumetric medical images presents a significant challenge, particularly in modeling spatial dependencies across different anatomical planes. Although 3D fully convolutional networks (FCNs) are widely used, the effectiveness of spatial feature extraction in such architectures may remain limited. In this study, a method called 3D Rubik Convolution, which applies 3D convolutions independently along the transaxial, coronal, and sagittal planes, is proposed. Unlike conventional convolution approaches that process the entire volume from a single perspective, the proposed method aims to extract spatial information separately across multiple anatomical planes while preserving full 3D representation. The method was evaluated on a COVID-19 lung computed tomography (CT) dataset and demonstrated higher segmentation accuracy compared to a baseline FCN architecture. The obtained results indicate that multi-view convolutional strategies can enhance segmentation performance by more effectively modeling spatial relationships.