EvoBrainNet: A Multi-Objective Evolutionary Neural Architecture Search With Self-Adaptive Mutation for Volumetric Brain Tumor Segmentation.

Accurate brain tumor segmentation in volumetric magnetic resonance imaging (MRI) remains a significant challenge due to the heterogeneity in tumor shape, size, location, and intensity. To address this, we propose EvoBrainNet, a multi-objective evolutionary framework that optimizes both segmentation accuracy and model efficiency. We design a modular 3D search space comprising three core components: the ExoFeature module for enhanced contextual encoding, the DRAP (dilated residual attention pyramid) module for multiscale attention and channel recalibration, and the RefineUp module for decoder-side refinement via attention-guided upsampling. These modules are embedded into a supernet architecture supporting scalable and flexible architectural search. An evolutionary neural architecture search (NAS) with a self-adaptive mutation strategy enables dynamic exploration of architectures based on performance feedback. The search jointly optimizes segmentation quality, measured by dice similarity coefficient (DSC), and model efficiency, evaluated by parameter count and floating-point operations (GFLOPs). To the best of our knowledge, this is the first multi-objective evolutionary NAS framework employing self-adaptive mutation for brain tumor segmentation. Experiments on the BraTS 2021 dataset show that EvoBrainNet achieves state-of-the-art performance with a 95.74% DSC and 1.30 mm HD95, along with high IoU, precision, recall, and specificity. It outperforms twelve existing methods in both accuracy and efficiency. Moreover, it exhibits consistent performance across diverse datasets, obtaining 93.55% DSC on BraTS 2020, 94.35% DSC on MSD Brain Tumor, and 82.61% DSC on ACDC. Ablation and statistical analyses confirm the contribution of each module and the effectiveness of the adaptive mutation strategy.

Authors

Publication Details

Journal
PubMed
Published
2026-11-20
DOI
https://doi.org/10.1109/jbhi.2026.3727770
Primary Topic
Brain Tumor Detection and Classification
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

EvoBrainNet: A Multi-Objective Evolutionary Neural Architecture Search With Self-Adaptive Mutation for Volumetric Brain Tumor Segmentation.

Mahade Hasan, Thippa Reddy Gadekallu, Pronaya Bhattacharya, Farhana Yasmin et al.
PubMed
Brain Tumor Detection and Classification
article

EvoBrainNet: A Multi-Objective Evolutionary Neural Architecture Search With Self-Adaptive Mutation for Volumetric Brain Tumor Segmentation.

Mahade Hasan, Thippa Reddy Gadekallu, Pronaya Bhattacharya, Farhana Yasmin, Haipeng Liu, Yu Xue
article en

Abstract

Accurate brain tumor segmentation in volumetric magnetic resonance imaging (MRI) remains a significant challenge due to the heterogeneity in tumor shape, size, location, and intensity. To address this, we propose EvoBrainNet, a multi-objective evolutionary framework that optimizes both segmentation accuracy and model efficiency. We design a modular 3D search space comprising three core components: the ExoFeature module for enhanced contextual encoding, the DRAP (dilated residual attention pyramid) module for multiscale attention and channel recalibration, and the RefineUp module for decoder-side refinement via attention-guided upsampling. These modules are embedded into a supernet architecture supporting scalable and flexible architectural search. An evolutionary neural architecture search (NAS) with a self-adaptive mutation strategy enables dynamic exploration of architectures based on performance feedback. The search jointly optimizes segmentation quality, measured by dice similarity coefficient (DSC), and model efficiency, evaluated by parameter count and floating-point operations (GFLOPs). To the best of our knowledge, this is the first multi-objective evolutionary NAS framework employing self-adaptive mutation for brain tumor segmentation. Experiments on the BraTS 2021 dataset show that EvoBrainNet achieves state-of-the-art performance with a 95.74% DSC and 1.30 mm HD95, along with high IoU, precision, recall, and specificity. It outperforms twelve existing methods in both accuracy and efficiency. Moreover, it exhibits consistent performance across diverse datasets, obtaining 93.55% DSC on BraTS 2020, 94.35% DSC on MSD Brain Tumor, and 82.61% DSC on ACDC. Ablation and statistical analyses confirm the contribution of each module and the effectiveness of the adaptive mutation strategy.

PubMedVol. PP
Brain Tumor Detection and Classification
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon