CIPA

<div align="center"> <h1> Multi-Modal Interactive Perception Network with Mamba for Lung Tumor Segmentation in PET-CT Images </h1>

Jie Mei¹, Chenyu Lin², Yu Qiu³, Yaonan Wang¹, Hui Zhang¹, Ziyang Wang⁴, Dong Dai⁴</sup>

¹ Hunan University, ² Nankai University, ³ Hunan Normal University, ⁴ Tianjin Medical University Cancer Institute and Hospital

</div>

• This repository contains the official code for paper Multi-Modal Interactive Perception Network with Mamba for Lung Tumor Segmentation in PET-CT Images.
• This paper has been accepted to CVPR 2025.
• This code and PCLT20K dataset are licensed for non-commerical research purpose only.

Introduction

<p style="text-align:justify; text-justify:inter-ideograph;"> Lung cancer is a leading cause of cancer-related deaths globally. PET-CT is crucial for imaging lung tumors, providing essential metabolic and anatomical information, while it faces challenges such as poor image quality, motion artifacts, and complex tumor morphology. Deep learning-based models are expected to address these problems, however, existing small-scale and private datasets limit significant performance improvements for these methods. Hence, we introduce a large-scale PET-CT lung tumor segmentation dataset, termed PCLT20K, which comprises 21,930 pairs of PET-CT images from 605 patients. Furthermore, we propose a cross-modal interactive perception network with Mamba (CIPA) for lung tumor segmentation in PET-CT images. Specifically, we design a channel-wise rectification module (CRM) that implements a channel state space block across multi-modal features to learn correlated representations and helps filter out modality-specific noise. A dynamic cross-modality interaction module (DCIM) is designed to effectively integrate position and context information, which employs PET images to learn regional position information and serves as a bridge to assist in modeling the relationships between local features of CT images. Extensive experiments on a comprehensive benchmark demonstrate the effectiveness of our CIPA compared to the current state-of-the-art segmentation methods. We hope our research can provide more exploration opportunities for medical image segmentation. </p>

Environment

1. Create environment.

   conda create -n MIPA python=3.10
   conda activate MIPA
   

2. Install all dependencies. Install pytorch, cuda and cudnn, then install other dependencies via:

   pip install torch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 --index-url https://download.pytorch.org/whl/cu118
   

   pip install -r requirements.txt
   

3. Install selective_scan_cuda_core.

   cd models/encoders/selective_scan
   pip install .
   cd ../../..
   

PCLT20K

Please contact Jie Mei (jiemei AT hnu.edu.cn) for the dataset. We will get back to you shortly. The email should contain the following information. Note: For better academic communication, a real-name system is encouraged and your email suffix must match your affiliation (e.g., hello@hnu.edu.cn). If not, you need to explain why.

Name: (Tell us who you are.)
Affiliation: (The name/url of your institution or university, etc.)
Job Title: (E.g., Professor, Associate Professor, PhD, etc.)
Email: (Dataset will be sent to this email.)
How to use: (Only for non-commercial use.)

Data Preparation

1. For our dataset PCLT20K, we orgnize the dataset folder in the following structure:

   <PCLT20K>
       |-- <0001>
           |-- <name1_CT.png>
           |-- <name1_PET.png>
           |-- <name1_mask.png>
           ...
       |-- <0002>
           |-- <name2_CT.png>
           |-- <name2_PET.png>
           |-- <name2_mask.png>
           ...
       ...
       |-- train.txt
       |-- test.txt
   

   train.txt/test.txt contains the names of items in training/testing set, e.g.:

   <name1>
   <name2>
   ...
   

2. Please put our dataset in the data directory

Usage

Training

1. Please download the pretrained VMamba weights, and put them under pretrained/vmamba/. We use VMamba_Tiny as default.

   • VMamba_Tiny
   • VMamba_Small
   • VMamba_Base

2. Config setting.

   Edit config in the train.py. Change C.backbone to sigma_tiny / sigma_small / sigma_base to use the three versions of VMamba.

3. Run multi-GPU distributed training:

   torchrun --nproc_per_node 'GPU_Numbers' train.py
   

4. You can also use single-GPU training:

   python train.py
   

5. Results will be saved in save_model folder.

Testing

The pretrained model of CIPA (CIPA.pth) can be downloaded:

• Baidu Yunpan, Password: CIPA
• Google Drive

python pred.py

Citation

If you are using the code/model provided here in a publication, please consider citing:

@inproceedings{mei2025cross,
  title={Cross-Modal Interactive Perception Network with Mamba for Lung Tumor Segmentation in PET-CT Images},
  author={Mei, Jie and Lin, Chenyu and Qiu, Yu and Wang, Yaonan and Zhang, Hui and Wang, Ziyang and Dai, Dong},
  booktitle={IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2025}
}

Contact

For any questions, please contact me via e-mail: jiemei AT hnu.edu.cn.

Acknowledgment

This project is based on the VMamba and Sigma, thanks for their excellent works.