H2ASeg: Hierarchical Interaction and Weighting Network for Tumor Segmentation in PET/CT images

Paper

• This project is the open source code of H2ASeg

Usage:

• Datasets

  • Automated Lesion Segmentation in PET/CT Challenge
  • MICCAI Hecktor 2022 Challenge

• Train

python -u train.py

Code checklist for machine learning-based MICCAI papers

Environments and Requirements

• Ubuntu version: Ubuntu 20.04.6 LTS
• CPU: AMD EPYC 7763 64-Core Processor
• GPU: NVIDIA GeForce RTX 4090
• CUDA: 12.2
• python: 3.10.16

To install requirements:

pip install -r requirements.txt

Dataset

• Automated Lesion Segmentation in PET/CT Challenge
• MICCAI Hecktor 2022 Challenge

Preprocessing

A brief description of the preprocessing method

• registration
• intensity normalization

Running the data preprocessing code:

python registration.py
python preprocessing.py

Training

To train the model(s) in the paper, run this command:

python train.py

Inference and Evaluation

To infer the testing cases and compute the evaluation metrics, run this command:

python inference.py

Results

Our method achieves the following performance on Automated Lesion Segmentation in PET/CT Challenge and MICCAI Hecktor 2022 Challenge

Dateset name	Model name	DICE	95% Hausdorff Distance
AutoPET-II	H2ASeg	60.03%	63.09
Hecktor2022	H2ASeg	59.69%	131.92

H2ASeg_JinPLU

H2ASeg_JinPLU