mednext

[MICCAI 2023] MedNeXt is a fully ConvNeXt architecture for 3D medical image segmentation.

MIST: A simple and scalable end-to-end framework for 3D medical imaging segmentation.

Fully automated 3D perivascular space (PVS) segmentation tool for T1w and T2w brain MRI.

PyTorch implementation of MMA-UNet for automatic brain tumor segmentation from contrast-enhanced MRI using MedNeXt, Mamba and CBAM.

Precision Pays — fixed-weight Kervadec boundary loss (lambda=0.05, theta0=30) on a MedNeXt/nnU-Net backbone, BraTS-2023 adult glioma, 5-fold CV under the complete official metric set: a transparently reported negative result, and the best connected-component consensus veto. Paper EN+FR, code, exact result JSONs, figures, tables, dataset provenance.

The Gate Does Not Choose — an expert-initialised patch-wise MoE vs 24 training-free consensus operators (ordered vetoes, 1-vs-3 vetoes, symmetric votes), BraTS-2023 adult glioma, 5 folds under the official lesion-wise Dice/HD95. Paper EN+FR, code, exact result JSONs, figures, tables, dataset provenance.

Production-oriented MedNeXt models with checkpoint compatibility, selective activation checkpointing, and optional CUDA acceleration

Distance Map auxiliary loss for brain tumor segmentation (BraTS 2023 GLI) : characterisation of a topological fragments artefact + parameter-free CC-consensus filter improving HD95 NCR. Paper, code, reproducibility data.