Mamba-augmented 3D encoder–decoder with selective state-space bottleneck for segmentation
Use a 3D convolutional encoder–decoder segmentation backbone, inserting a Mamba-style selective state-space block specifically at the bottleneck to model global context while keeping local feature extraction in the convolutional parts.
Discovered 2026-11-10 · Sources checked 2026-11-10
Why it matters
For tasks like hippocampal segmentation across multiple MRI visits, placing the selective state-space module at the bottleneck is intended to improve segmentation quality in a way that supports downstream longitudinal measurements (e.g., volume ranking and change estimates).
What changed
A conventional 3D convolutional encoder–decoder is modified by adding a Mamba selective state-space block at the bottleneck, and the network is designed to ingest one or more MRI visits of a subject.
Where it helps
- Hippocampal segmentation from 3D MRI for Alzheimer's-related structural measures
- Longitudinal segmentation across repeated MRI visits with downstream volumetry
- Other 3D medical image segmentation problems needing stronger bottleneck context modeling
Evidence & reproduction
Reproduction not evaluated. A linked paper or repository does not establish independent reproduction.
Limitations & open questions
The cited results are tied to specific datasets, validation protocols, and evaluation settings; absolute volume agreement with a reference tool is described as systematically smaller, and some distance metrics are reported as not significantly different from certain baselines.
Sources
- HippoMamba: A Mamba-augmented network for hippocampal segmentation and serial volumetry in Alzheimers disease ↗
arXiv · primary · Checked 2026-11-10
