Post

New research · Oncology
Journal of imaging · 1w
Cohort studyJournal of imaging · 2026

GDNet: A Robust 2.5D Multimodal MRI Brain Tumor Segmentation Framework with EMA Stabilization and Tumor-Aware Sampling.

Behnam Kiani Kalejahi, Sajid Khan, Mohammad Javad Rajabi
Read paper
OncologyCohort study

A 2.5D deep-learning model segmented whole brain tumor on MRI with 0.791 Dice accuracy

GDNet: A Robust 2.5D Multimodal MRI Brain Tumor Segmentation Framework with EMA Stabilization and Tumor-Aware Sampling.

Behnam Kiani Kalejahi … Mohammad Javad Rajabi
Journal of imaging · 2026
Background

Accurate, automated delineation of adult diffuse gliomas from multi-parametric magnetic resonance imaging (mpMRI) is central to quantitative neuro-oncology.

Methods

We introduce GDNet, a 2.5D multimodal MRI segmentation framework for adult glioma evaluated on the BraTS 2024 cohort.

Results

Automated whole-tumor outlines overlapped expert outlines fairly well (1.0 = perfect match)

WT
0.791Dice
TC
0.736Dice
ET
0.654Dice
More results

Volumetric 3D networks dominate the BraTS leaderboard but require expensive GPUs, long training cycles, and provide diminishing returns relative to their compute budget.

Slice-wise 2D models, by contrast, discard inter-slice context that is informative for thin tumor rims and small enhancing foci.

More results

Validation positive-only scores were 0.805 ± 0.002 (whole tumor), 0.757 ± 0.004 (tumor core), 0.683 ± 0.009 (enhancing tumor).

“
Conclusion · 1 of 2

Carefully engineered training strategies, tumor-aware sampling, Exponential Moving Average stabilization, and a modest 2.5D context window recover a substantial fraction of the accuracy of much heavier 3D networks at a fraction of the compute, are reproducible across seeds, and outperform a heavier GDNet-inspired architectural variant on the same data.

Conclusion · 2 of 2

GDNet is therefore a practical and, pending external validation, potentially clinically deployable framework for multimodal glioma segmentation on workstation-class GPU hardware.

Read paper
0 comments

No comments yet. Be the first.

Related papers

LatestFoundational
AI / Informatics
1.16%
Safety Agent gating reduced incorrect patient inclusion to 1.16% for non-small cell lung cancer trials.
Guideline
at least 20years
patients need this much follow-up because the cancer can come back many years later
AI / Informatics
32
studies exploring machine learning to predict cancer outcomes
Randomized Trial
hazard ratio 0·67
pirarubicin instillation led to a lower risk of recurrence compared to observation
Guideline
“As more new targeted agents become available and hypomethylating agent-based triplet combinations are used more widely, we fully anticipate that we will, in some years' time, write a fourth guideline on the management of AML in older patients.”
AI / Informatics
AUC = 0.939
This shows the model's excellent ability to predict ovarian cancer treatment response
Guideline
50 and 70 years of age
Australians within this age range are eligible for lung cancer screening
Case Report
12 mm
the brain's center was pushed sideways due to severe swelling