3D-ARUNet:3D Attention-Based Residual U-Net with ASPP for Multimodal MRI Brain Tumor Segmentation

View at publisher Summarise with VIRAT
Authors: Noreen Fatima, Yassine Himeur, Hamzah Ali Alkhazaleh, Shadi Atalla, Murad Almadani, Wathiq Mansoor, Fayçal Bensaali
Cited by: 0
References: 17

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Accurately outlining brain tumors on MRI matters guides treatment planning, tracks response, and supports prognosis. Yet even modern models struggle because tumors look different from one patient to the next, their borders blend with healthy tissue, and MRI intensities overlap—issues that often produce fuzzy boundaries and poor generalization. We introduce 3D-ARUNet, a 3D U-Net variant designed to tackle these pain points. It combines residual double-convolution blocks (to keep gradients stable and learning efficient), attention-gated skip connections (to quiet background clutter and emphasize tumor cues), and a lightweight ASPP module at the bottleneck (to capture multi-scale context without heavy compute). In plain terms, the network learns richer features, looks where it matters, and understands structure at multiple sizes all at once. Using the BraTS dataset with multimodal MRI (FLAIR, T1ce, T2) and expert annotations, 3D-ARUNet delivered strong, consistent performance across tumor subregions. We observed mean Dice scores of 0.871 ± 0.068 for edema (ED), 0.811 ± 0.091 for the necrotic/non-enhancing core (NCR/NET), and 0.843 ± 0.090 for the enhancing tumor (ET), indicating precise, clinically useful segmentations.

Fields of Study & Subjects

Brain Tumor Detection and Classification Medical Image Segmentation Techniques Glioma Diagnosis and Treatment

Publication details

Journal / source
International Conference on Consumer Electronics
Pages
1–6
Publication date
3 February 2026
Language
EN