JCBTD (Jun Cheng Brain Tumor Dataset) — T1-Weighted Contrast-Enhanced Brain MRI with Tumor Segmentation

This dataset is derived from the Jun Cheng Brain Tumor Dataset (JCBTD), a small-scale collection of T1-weighted contrast-enhanced brain MRI images with pathologically confirmed brain tumors, originally hosted on Figshare. The original collection provides brain MRI images in MATLAB format (.mat) with manually delineated tumor masks. Here, we converted the original MATLAB files into PNG images and binary tumor masks. The dataset is designed for brain tumor segmentation, and classification.

📦 Dataset Structure

The dataset has a single train split, with each row corresponding to a single 2D axial slice from a patient's T1-weighted contrast-enhanced MRI:

Field	Description
slice_id	Unique slice identifier (e.g., 1, 2, ..., 3064)
image	2D T1-weighted contrast-enhanced MRI slice (512 × 512 pixels, normalized to uint8)
mask	2D binary tumor segmentation mask (512 × 512 pixels; 255 = tumor region, 0 = background)
label	Tumor type: 1 (meningioma), 2 (glioma), 3 (pituitary tumor)

Class Distribution:

• Meningioma: 708 slices
• Glioma: 1,426 slices
• Pituitary tumor: 930 slices
• Total: 3,064 slices from 233 patients

⚙️ Preprocessing

• Original MATLAB (.mat) files were loaded using h5py; each file contains a cjdata struct with image, tumor mask, and label fields
• Images were normalized to float32, rescaled to [0, 1], and converted to uint8 [0, 255] for storage as PNG
• Tumor masks were binarized (values > 0 → 255, all others → 0) to create binary segmentation masks
• All 3,064 slices were preserved
• Slice IDs correspond to the original MATLAB filename integers

🚀 Usage

from datasets import load_dataset
import matplotlib.pyplot as plt
import numpy as np

ds = load_dataset("chehablab/JCBTD", split="train")

# Grab a sample slice and view image with mask overlay
sample = ds[0]

fig, axes = plt.subplots(1, 3, figsize=(14, 4))

# Original image
axes[0].imshow(sample["image"], cmap="gray")
axes[0].set_title("T1-Weighted MRI")
axes[0].axis("off")

# Tumor mask
axes[1].imshow(sample["mask"], cmap="gray")
axes[1].set_title("Tumor Mask")
axes[1].axis("off")

# Image with mask overlay
axes[2].imshow(sample["image"], cmap="gray")
axes[2].imshow(sample["mask"], cmap="jet", alpha=0.4)
axes[2].set_title("MRI + Mask Overlay")
axes[2].axis("off")

# Tumor classification metadata
tumor_labels = {1: "Meningioma", 2: "Glioma", 3: "Pituitary Tumor"}
info_text = (
    f"Slice ID: {sample['slice_id']}\n"
    f"Tumor Type: {tumor_labels[sample['label']]}\n"
    f"Patient ID: {sample['patient_id']}"
)
fig.suptitle(info_text, fontsize=10)
plt.tight_layout()
plt.show()

# Filter by tumor type
meningioma_slices = ds.filter(lambda x: x["label"] == 1)
print(f"Meningioma slices: {len(meningioma_slices)}")

---

🏷️ Feature Descriptions

Tumor Classification

Three distinct histopathological subtypes of primary brain tumors:

• Meningioma (label=1): 708 slices — Benign tumor arising from meninges (brain membranes); typically slow-growing with favorable prognosis
• Glioma (label=2): 1,426 slices — Malignant tumor arising from glial cells; includes high-grade aggressive variants with worse prognosis
• Pituitary Tumor (label=3): 930 slices — Tumor of the pituitary gland; often benign but may cause endocrine dysfunction

Image Specifications

• Modality: T1-weighted contrast-enhanced MRI (post-gadolinium DTPA injection)
• In-plane resolution: 512 × 512 pixels
• Pixel dimensions: 0.49 × 0.49 mm²
• Slice thickness: 6 mm
• Slice gap: 1 mm
• Gadolinium dose: 0.1 mmol/kg at 2 ml/s injection rate

Acquisition Protocol

All images were acquired at:

• Nanfang Hospital, Guangzhou, China
• General Hospital, Tianjin Medical University, China
• Time period: September 2005 – October 2010

---

🧪 Use Cases

• Brain tumor segmentation from T1-weighted contrast-enhanced MRI
• Multi-class tumor classification (meningioma, glioma, pituitary)
• Tumor region detection and localization
• Representation learning and feature extraction for medical imaging
• Self-supervised pretraining on brain MRI
• Comparison of segmentation performance across tumor subtypes
• Data augmentation studies on tumor regions

⚠️ Important Notes

• Class imbalance: Glioma is more prevalent (1,426 slices) compared to meningioma (708) and pituitary tumor (930). Use weighted sampling or class balancing as needed.
• Tumor annotation: Tumor masks are binary; voxels with value 255 indicate the tumor region as manually delineated by clinical experts. Background is 0.
• Image normalization: Images were normalized using min-max scaling (per-image); further normalization may be needed depending on your application.

📚 Citation

This dataset is derived from data originally published on Figshare under a CC BY 4.0 license. If you use this dataset, please aknowledge [chehablab.com](Chehab Lab) cite the original Jun Cheng Brain Tumor Dataset:

@misc{Cheng2015JCBTD,
  author = {Cheng, Jun},
  title  = {Brain Tumor Dataset [dataset]},
  year   = {2015},
  publisher = {Figshare},
  doi    = {10.6084/m9.figshare.1512427}
}

Additionally, cite the research papers that introduced this dataset:

@article{Cheng2015Enhanced,
  author = {Cheng, Jun and others},
  title  = {Enhanced Performance of Brain Tumor Classification via Tumor Region Augmentation and Partition},
  journal = {PLOS One},
  volume = {10},
  number = {10},
  year = {2015}
}

@article{Cheng2016Retrieval,
  author = {Cheng, Jun and others},
  title  = {Retrieval of Brain Tumors by Adaptive Spatial Pooling and Fisher Vector Representation},
  journal = {PLOS One},
  volume = {11},
  number = {6},
  year = {2016}
}

✉️ Original Author Contact

Jun Cheng
School of Biomedical Engineering
Shenzhen University, Shenzhen, China
Email: chengjun583@qq.com

GitHub: https://github.com/chengjun583/brainTumorRetrieval (MATLAB source codes and additional resources)

📜 License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license, consistent with the original Jun Cheng Brain Tumor Dataset.

You may copy, modify, distribute, and use the data for any purpose (commercial or non-commercial), provided that appropriate credit is given to Jun Cheng, Figshare, and the original papers.

Chehab Lab @ 2026