Introduction

This repository contains the code for our paper:
XFMamba: Cross-Fusion Mamba for Multi-View Medical Image Classification
[Paper on arXiv] — Accepted at MICCAI 2025.

---

Environment

1. Create a new environment:

   conda create -n xfmamba python=3.10
   conda activate xfmamba
   

2. Install dependencies:

   pip install torch==2.2 torchvision torchaudio triton pytest chardet yacs termcolor fvcore seaborn packaging ninja einops numpy==1.24.4 timm==0.4.12
   

3. Install Mamba:

   pip install https://github.com/state-spaces/mamba/releases/download/v2.2.4/mamba_ssm-2.2.4+cu12torch2.2cxx11abiTRUE-cp310-cp310-linux_x86_64.whl
   

---

Datasets

We use the following three publicly available medical imaging datasets:

• MURA
• CheXpert
• CBIS-DDSM

The dataset CSV:

• MURA
• CheXpert
• CBIS-DDSM

---

Training and Inference

1. Download the pre-trained VMamba weights and place them in the pretrained/vmamba directory:

   • VMamba_Tiny
   • VMamba_Small
   • VMamba_Base

2. Train the model:

   python 1_train_model.py \
       --root_dir '.' \
       --dataset 'mura' \
       --model_name 'dualfusionmambav13' \
       --num_tasks 1 \
       --num_classes 2 \
       --confusion_matrix_folder '.' \
       --savemodel_path '.' \
       --train_num 1
   

3. Run inference:

   python 2_inference_mura.py \
       --model_name 'dualfusionmambav13' \
       --model_path ''
   

---

Results

XFMamba achieves a strong balance between performance and efficiency, offering a lightweight and high-performing solution for multi-view medical image classification.

FLOPs and AUROC results are based on the CBIS-DDSM dataset.

MURA (2 Classes)

Architecture	Backbone	AUROC	Weight Link
XFMamba	VMamba-T	89.8%	XFMamba-T
XFMamba	VMamba-S	91.0%	XFMamba-S
XFMamba	VMamba-B	90.4%	XFMamba-B

CheXpert (13 Classes)

Architecture	Backbone	AUROC	Weight Link
XFMamba	VMamba-T	91.7%	XFMamba-T
XFMamba	VMamba-S	91.8%	XFMamba-S
XFMamba	VMamba-B	91.9%	XFMamba-B

CBIS-DDSM (2 Classes)

Architecture	Backbone	AUROC	Weight Link
XFMamba	VMamba-T	66.4%	XFMamba-T
XFMamba	VMamba-S	75.2%	XFMamba-S
XFMamba	VMamba-B	76.1%	XFMamba-B

---

Citation

If you find this repository helpful for your research, please consider citing our work:

@inproceedings{zheng2025xfmamba,
  title={Xfmamba: Cross-fusion mamba for multi-view medical image classification},
  author={Zheng, Xiaoyu and Chen, Xu and Gong, Shaogang and Griffin, Xavier and Slabaugh, Greg},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={672--682},
  year={2025},
  organization={Springer}
}