wlsdzyzl/MedPointS-cls

This repository contains the MedPointS-CLS dataset, a large-scale medical point cloud dataset for anatomy classification, completion, and segmentation, as presented in the paper Hierarchical Feature Learning for Medical Point Clouds via State Space Model.

• Paper: Hierarchical Feature Learning for Medical Point Clouds via State Space Model
• Project Page: https://flemme-docs.readthedocs.io/en/latest/medpoints.html
• Code: https://github.com/wlsdzyzl/flemme

Abstract

Deep learning-based point cloud modeling has been widely investigated as an indispensable component of general shape analysis. Recently, transformer and state space model (SSM) have shown promising capacities in point cloud learning. However, limited research has been conducted on medical point clouds, which have great potential in disease diagnosis and treatment. This paper presents an SSM-based hierarchical feature learning framework for medical point cloud understanding. Specifically, we down-sample input into multiple levels through the farthest point sampling. At each level, we perform a series of k-nearest neighbor (KNN) queries to aggregate multi-scale structural information. To assist SSM in processing point clouds, we introduce coordinate-order and inside-out scanning strategies for efficient serialization of irregular points. Point features are calculated progressively from short neighbor sequences and long point sequences through vanilla and group Point SSM blocks, to capture both local patterns and long-range dependencies. To evaluate the proposed method, we build a large-scale medical point cloud dataset named MedPointS for anatomy classification, completion, and segmentation. Extensive experiments conducted on MedPointS demonstrate that our method achieves superior performance across all tasks. The dataset is available at this https URL . Code is merged to a public medical imaging platform: this https URL .

MedPointS-CLS

This is the medical point cloud classification dataset from MedPointS, where data is input point cloud, and label is the class label.

Each point cloud has been normalized and sub-sampled to 2048 points. The correspondence between class names and labels is listed as follows (the label value plus 1 is the actual key of following map):

coarse_label_to_organ = {1: 'adrenalgland',
    2: 'aorta',
    3: 'autochthon',
    4: 'bladder',
    5: 'brain',
    6: 'breast',
    7: 'bronchie',
    8: 'celiactrunk',
    9: 'cheek',
    10: 'clavicle',
    11: 'colon',
    12: 'costa',
    13: 'duodenum',
    14: 'esophagus',
    15: 'eyeball',
    16: 'femur',
    17: 'gallbladder',
    18: 'gluteusmaximus',
    19: 'heart',
    20: 'hip',
    21: 'humerus',
    22: 'iliacartery',
    23: 'iliacvena',
    24: 'iliopsoas',
    25: 'inferiorvenacava',
    26: 'kidney',
    27: 'liver',
    28: 'lung',
    29: 'mediastinaltissue',
    30: 'pancreas',
    31: 'portalveinandsplenicvein',
    32: 'smallbowel',
    33: 'spleen',
    34: 'stomach',
    35: 'thymus',
    36: 'thyroid',
    37: 'trachea',
    38: 'uterocervix',
    39: 'uterus',
    40: 'vertebrae',
    41: 'gonads',
    42: 'sacrum',
    43: 'clavicula',
    # 44: 'prostate',
    44: 'pulmonaryartery',
    # 45: 'ribcartilage',
    45: 'rib',
    46: 'scapula',
    # 48: 'skull',
    # 49: 'spinalcanal',
    # 50: 'sternum'
    }

Sample Usage

The associated code repository flemme provides clear instructions and configuration files for training and testing models on the MedPointS dataset for classification, completion, and segmentation tasks.

To train and evaluate models, first set up the flemme environment (refer to the Github README for installation instructions). Then, use the following commands:

## classification
train_flemme --config /path/to/project/flemme/resources/pcd/medpoints/cls/train_pointmamba2knn_clm.yaml
test_flemme --config /path/to/project/flemme/resources/pcd/medpoints/cls/test_pointmamba2knn_clm.yaml
## completion
train_flemme --config /path/to/project/flemme/resources/pcd/medpoints/cpl/train_pointmamba2knn_cpl.yaml
test_flemme --config /path/to/project/flemme/resources/pcd/medpoints/cpl/test_pointmamba2knn_cpl.yaml
## segmentation
train_flemme --config /path/to/project/flemme/resources/pcd/medpoints/seg/train_pointmamba2knn_sem.yaml
test_flemme --config /path/to/project/flemme/resources/pcd/medpoints/seg/test_pointmamba2knn_sem.yaml

If you find our project helpful, please consider to cite the following works:

@misc{zhang2025hierarchicalfeaturelearningmedical,
      title={Hierarchical Feature Learning for Medical Point Clouds via State Space Model}, 
      author={Guoqing Zhang and Jingyun Yang and Yang Li},
      year={2025},
      eprint={2504.13015},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2504.13015}, 
}

---

dataset_info: features: - name: data sequence: sequence: float32 - name: label sequence: float32 splits: - name: train num_bytes: 947171520 num_examples: 28737 download_size: 718817756 dataset_size: 947171520 configs: - config_name: default data_files: - split: train path: data/train-*