Xiang-zx-zx/CAR-T_NK_dataset

CAR-T/NK Immunological Synapse Dataset

Dataset Description

This dataset contains fluorescence microscopy images of Chimeric Antigen Receptor (CAR)-T and NK cell immunological synapses (IS) with corresponding instance segmentation masks. The images were acquired from patient-derived CAR-T/NK cell samples stained with multiple fluorescent markers including Perforin, P-Zeta, F-actin, and IgG.

The dataset is designed for training and evaluating object detection and instance segmentation models on biomedical cell microscopy images.

Dataset Structure

data_hf/
├── images/              # 156 fluorescence microscopy images (full dataset)
├── masks/               # 156 instance segmentation masks (full dataset)
├── train/               # 93 training images
├── train_annotation/    # 93 training masks
├── test/                # 31 test images
├── test_annotation/     # 31 test masks
├── val/                 # 32 validation images
└── val_annotation/      # 32 validation masks

Dataset Statistics

Split	Images	Masks
Train	93	93
Test	31	31
Validation	32	32
Total	156	156

Image Properties

• Resolution: 1024 x 1024 pixels
• Format: PNG
• Modality: Multi-channel fluorescence microscopy
• Fluorescent markers: AF647, AF488 (Perforin), AF568 (P-Zeta), AF405 (F-actin), and others depending on CAR construct

Annotation Details

Each mask is an instance segmentation annotation where individual cells are labeled with unique pixel values. The masks enable both bounding box detection and pixel-level instance segmentation tasks.

Usage

from datasets import load_dataset

dataset = load_dataset("YOUR_USERNAME/cart-nk-is-dataset")

Or directly load images:

from PIL import Image
import os

image = Image.open("train/example.png")
mask = Image.open("train_annotation/example.png")

Citation

If you use this dataset, please cite:

@article{zhang2025data,
  title={Data Augmentation for High-Fidelity Generation of CAR-T/NK Immunological Synapse Images},
  author={Zhang, Xiang and Zhang, Boxuan and Naghizadeh, Alireza and Mohamed, Mohab and Liu, Dongfang and Tang, Ruixiang and Metaxas, Dimitris and Liu, Dongfang},
  journal={Frontiers},
  year={2025}
}

License

This dataset is released under the CC BY 4.0 license.