View in EDS

MDS-ResUNet: A Multi-scale Deeply Supervised Framework for Precise Brain Tumor Segmentation with Enhanced Boundary Delineation.

Saved in:
Bibliographic Details
Title: MDS-ResUNet: A Multi-scale Deeply Supervised Framework for Precise Brain Tumor Segmentation with Enhanced Boundary Delineation.
Authors: Hussain, Ashfaq1 (AUTHOR) ashfaq21_rs@ece.nits.ac.in, Laskar, Rabul Hussain1 (AUTHOR), Bhuyan, Manas Kamal2 (AUTHOR)
Source: Arabian Journal for Science & Engineering (Springer Science & Business Media B.V. ). Mar2026, Vol. 51 Issue 6, p7619-7640. 22p.
Subjects: Image segmentation, Magnetic resonance imaging, Feature extraction, Edge detection (Image processing), Convolutional neural networks, Artificial neural networks, Deep learning, Diagnostic imaging
Abstract: Brain tumor segmentation from multimodal MRI scans is still a hard and crucial problem in medical imaging, and the outcome directly affects diagnosis, treatment plan, and patient prognosis. Current deep learning models like U-Net and its traditional variations tend to be limited in detecting fine-grained tumor boundaries and modeling multi-scale contextual information, especially when dealing with heterogeneous tumor structure and contrast-poor areas. These shortcomings result in poor segmentation performance, particularly in defining intricate tumor areas. To remedy these issues, we introduced MDS-ResUNet, an improved U-Net-inspired architecture that is particularly designed to enhance brain tumor segmentation precision. The model presented a few major innovations: (1) multi-scale residual learning, which enabled hierarchical feature aggregation across decoder levels to maintain semantic coherence and spatial resolution; (2) an atrous pyramid pooling (APP) module, enabling multi-scale contextual feature extraction by dilated convolutions, strengthening the model to detect tumors with different sizes and morphologies; (3) an edge-aware module (EAM), enabling tumor boundary defining by promoting edge feature representations explicitly; and (4) deep supervision, enforced on multiple decoder levels to encourage stable gradient flow and efficient multi-scale feature learning. Experimental verification was performed on the BraTS19 dataset, wherein MDS-ResUNet showed better performance with Dice values of 0.946 (Whole Tumor), 0.903 (Tumor Core), and 0.843 (Enhancing Tumor). Competitive IoU, sensitivity, and specificity measures were also reported. Through the proper resolution of the shortcomings of existing methods, MDS-ResUNet created a new state of the art in brain tumor segmentation, showing promising future applications in both deep learning research and clinical use. [ABSTRACT FROM AUTHOR]
Copyright of Arabian Journal for Science & Engineering (Springer Science & Business Media B.V. ) is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:Brain tumor segmentation from multimodal MRI scans is still a hard and crucial problem in medical imaging, and the outcome directly affects diagnosis, treatment plan, and patient prognosis. Current deep learning models like U-Net and its traditional variations tend to be limited in detecting fine-grained tumor boundaries and modeling multi-scale contextual information, especially when dealing with heterogeneous tumor structure and contrast-poor areas. These shortcomings result in poor segmentation performance, particularly in defining intricate tumor areas. To remedy these issues, we introduced MDS-ResUNet, an improved U-Net-inspired architecture that is particularly designed to enhance brain tumor segmentation precision. The model presented a few major innovations: (1) multi-scale residual learning, which enabled hierarchical feature aggregation across decoder levels to maintain semantic coherence and spatial resolution; (2) an atrous pyramid pooling (APP) module, enabling multi-scale contextual feature extraction by dilated convolutions, strengthening the model to detect tumors with different sizes and morphologies; (3) an edge-aware module (EAM), enabling tumor boundary defining by promoting edge feature representations explicitly; and (4) deep supervision, enforced on multiple decoder levels to encourage stable gradient flow and efficient multi-scale feature learning. Experimental verification was performed on the BraTS19 dataset, wherein MDS-ResUNet showed better performance with Dice values of 0.946 (Whole Tumor), 0.903 (Tumor Core), and 0.843 (Enhancing Tumor). Competitive IoU, sensitivity, and specificity measures were also reported. Through the proper resolution of the shortcomings of existing methods, MDS-ResUNet created a new state of the art in brain tumor segmentation, showing promising future applications in both deep learning research and clinical use. [ABSTRACT FROM AUTHOR]
ISSN:2193567X
DOI:10.1007/s13369-025-10443-z