Swin-EVDNet v2 3D — Brain Tumour Segmentation

Research model. Not for clinical use.

Model description

3D Swin-Transformer encoder with evidential segmentation and burden classification, trained on the BraTS 2023 Adult Glioma dataset for 25 epochs.

• Input: 4 MRI channels (T1, T1ce, T2, FLAIR), shape (B, 4, D, H, W)
• Output: segmentation probability + uncertainty + 3-class burden classification
• Inference: 3D Gaussian-weighted sliding window, 96³ crops, 50% overlap

Intended use

Research prototype only. Not for medical diagnosis.

How to use

import torch
from huggingface_hub import hf_hub_download
from model import SwinEVDNet3D

path = hf_hub_download("aryan09103/swin-evdnet", "best_slim.pt")
model = SwinEVDNet3D(num_classes=3, encoder="swin")
state = torch.load(path, map_location="cpu", weights_only=False)
model.load_state_dict(state, strict=True)
model.eval()