LFC Dataset: Fetal Cerebellum 3D MRI Landmark Dataset

The LFC Dataset is a curated 3D fetal brain MRI dataset designed for fetal cerebellum landmark detection and fetal cerebellum biometric measurement. It supports research in fetal brain development, fetal cerebellum morphology, medical landmark detection, and 3D medical image analysis.

The dataset provides reconstructed 3D fetal MRI volumes with expert-reviewed fetal cerebellum landmark annotations. It can be used to train and evaluate landmark detection models such as the APGA network, as well as boundary-aware and domain-generalized medical landmark detection methods.

If you use this dataset, please cite the related papers listed in the Citation section.

---

📌 Related Papers

This dataset is associated with the following works:

1. Fetal Cerebellum Landmark Detection Based on 3D MRI: Method and Benchmark
   Haifan Gong*, Huixian Liu*, Yitao Wang, Xiaoling Liu, Xiang Wan, Qiao Shi, and Haofeng Li.
   IEEE Journal of Biomedical and Health Informatics, 2025.

2. Boundary as the Bridge: Toward Heterogeneous Partially-Labeled Medical Image Segmentation and Landmark Detection
   Haifan Gong, Boyao Wan, Luoyao Kang, Xiang Wan, Lingyan Zhang, and Haofeng Li.
   IEEE Transactions on Medical Imaging, 2025.

3. Domain Generalized Medical Landmark Detection via Robust Boundary-Aware Pre-Training
   Haifan Gong, Yu Lu, Xiang Wan, and Haofeng Li.
   AAAI Conference on Artificial Intelligence, 2025.

---

📦 Dataset Contents

The dataset includes:

• 300 de-identified fetal MRI cases
• 3D reconstructed fetal brain MRI volumes
• Six fetal cerebellum landmarks per case
• Gestational age for each subject
• Official train / validation / test split
• Preprocessing and reconstruction pipeline
• Training and evaluation code

A CSV file gestational_age_samples.csv is provided, containing gestational age and split information for each subject.

Column	Description
Sample_ID	Case identifier
Gestational_Age_Weeks	Gestational age in weeks
Split	Train / Validation / Test split

---

🧠 Landmark Definition

Each MRI volume contains six fetal cerebellum landmarks, corresponding to the two endpoints of three fetal cerebellum biometric measurements:

1. TCD1 / TCD2 — endpoints of the Transverse Cerebellar Diameter (TCD)
2. HDV1 / HDV2 — endpoints of the Height Diameter of the Vermis (HDV)
3. ADV1 / ADV2 — endpoints of the Anteroposterior Diameter of the Vermis (ADV)

These landmarks are used for:

• Fetal cerebellum biometric measurement
• Fetal brain and cerebellum development analysis
• 3D medical landmark detection benchmarking
• Evaluation of landmark detection robustness and generalization
• Training and evaluating models such as APGA and domain-generalized landmark detection methods

---

🏥 MRI Acquisition and Reconstruction

The original fetal MR scans were acquired as multi-view thick-slice fetal brain MRI, including axial, sagittal, and coronal T2-weighted single-shot Fast Spin Echo scans.

The original scans were reconstructed into isotropic 3D fetal brain MRI volumes. The reconstructed 3D volumes have the following spatial resolution:

Item	Value
Reconstructed spacing	0.6 × 0.6 × 0.6 mm³
Original slice thickness	5 mm
Original in-plane resolution	approximately 0.68 × 0.68 mm²

---

✍️ Annotation Protocol

The fetal cerebellum landmarks were first annotated by a radiologist and then checked and revised by another experienced radiologist to reach consensus.

The annotations include six anatomical landmarks per case, corresponding to the two endpoints of TCD, HDV, and ADV. These annotations can support direct landmark regression, heatmap-based landmark detection, and auxiliary pseudo-label segmentation used in boundary-aware frameworks.

---

📊 Dataset Statistics

Item	Description
Modality	Fetal brain MRI
Sequence	T2-weighted fetal MRI
Number of cases	300
Gestational age range	24–38 weeks
Cases per gestational week	20
Official split	120 train / 60 validation / 120 test
Landmark annotations	6 landmarks per case
Biometric measurements	TCD, HDV, ADV
Image format	NIfTI (.nii.gz)
Main task	3D fetal cerebellum landmark detection

---

📁 Suggested File Structure

A typical dataset directory may be organized as follows:

LFC_Dataset/
├── images/
│   ├── case_0001.nii.gz
│   ├── case_0002.nii.gz
│   └── ...
├── landmarks/
│   ├── case_0001.csv
│   ├── case_0002.csv
│   └── ...
├── gestational_age_samples.csv
├── train.txt
├── val.txt
├── test.txt
└── README.md

Each landmark annotation file may contain the 3D coordinates of the six fetal cerebellum landmarks:

landmark_name,x,y,z
TCD1,...
TCD2,...
HDV1,...
HDV2,...
ADV1,...
ADV2,...

---

🚨 Note on Benchmark Usage

The full LFC dataset contains 300 cases with an official 120 / 60 / 120 train-validation-test split.

In the domain-generalized medical landmark detection setting introduced in the AAAI 2025 paper, LFC is used as an external validation/testing domain, with 60 validation cases and 120 testing cases. Therefore, if you use LFC in the domain-generalization benchmark setting, please clearly distinguish this setting from the full LFC benchmark split.

---

✅ Intended Use

This dataset is intended for academic and non-commercial research on:

• Fetal brain MRI analysis
• Fetal cerebellum landmark detection
• Medical image landmark detection
• Medical image segmentation and landmark joint learning
• Domain generalization in medical image analysis
• Fetal biometric measurement and development analysis

---

⚠️ Limitations

Users should be aware of the following limitations:

• The dataset focuses on fetal cerebellum landmarks and is not intended for general fetal brain diagnosis.
• The dataset contains reconstructed 3D fetal MRI volumes derived from thick-slice clinical fetal MRI scans.
• Model performance may vary across scanners, acquisition protocols, gestational ages, and clinical sites.
• Any clinical use requires independent validation and regulatory approval.

---

📚 Citation

If you use this dataset, please cite the following papers:

@article{gong2025fetal,
  title   = {Fetal Cerebellum Landmark Detection Based on 3D {MRI}: Method and Benchmark},
  author  = {Gong, Haifan and Liu, Huixian and Wang, Yitao and Liu, Xiaoling and Wan, Xiang and Shi, Qiao and Li, Haofeng},
  journal = {IEEE Journal of Biomedical and Health Informatics},
  volume  = {29},
  number  = {8},
  pages   = {5712--5721},
  year    = {2025},
  doi     = {10.1109/JBHI.2025.3559702}
}

@article{gong2025boundary,
  title   = {Boundary as the Bridge: Toward Heterogeneous Partially-Labeled Medical Image Segmentation and Landmark Detection},
  author  = {Gong, Haifan and Wan, Boyao and Kang, Luoyao and Wan, Xiang and Zhang, Lingyan and Li, Haofeng},
  journal = {IEEE Transactions on Medical Imaging},
  volume  = {44},
  number  = {7},
  pages   = {2747--2756},
  year    = {2025},
  doi     = {10.1109/TMI.2025.3548919}
}

@inproceedings{gong2025domain,
  title     = {Domain Generalized Medical Landmark Detection via Robust Boundary-Aware Pre-Training},
  author    = {Gong, Haifan and Lu, Yu and Wan, Xiang and Li, Haofeng},
  booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
  volume    = {39},
  number    = {3},
  pages     = {3140--3148},
  year      = {2025},
  doi       = {10.1609/aaai.v39i3.32323}
}

---

License

This dataset is released under the Apache-2.0 License.

Please check the dataset repository and related papers for any additional usage requirements.