Image Classification
Transformers
Safetensors
Flair
vit
medical-imaging
brain-tumor
mri
vision-transformer
Instructions to use Songline/BrainTumor_FlairClassifier with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Songline/BrainTumor_FlairClassifier with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-classification", model="Songline/BrainTumor_FlairClassifier") pipe("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/hub/parrots.png")# Load model directly from transformers import AutoImageProcessor, AutoModelForImageClassification processor = AutoImageProcessor.from_pretrained("Songline/BrainTumor_FlairClassifier") model = AutoModelForImageClassification.from_pretrained("Songline/BrainTumor_FlairClassifier", device_map="auto") - Flair
How to use Songline/BrainTumor_FlairClassifier with Flair:
from flair.models import SequenceTagger tagger = SequenceTagger.load("Songline/BrainTumor_FlairClassifier") - Notebooks
- Google Colab
- Kaggle
Brain Tumor FLAIR Classifier
基于单份三维 FLAIR NIfTI 的脑肿瘤病例级二分类模型
模型基于 google/vit-base-patch16-224-in21k 微调
输入 .nii 或 .nii.gz
输出 yes_probability 与 yes 或 no 分类结果
Model Details
| 项目 | 内容 |
|---|---|
| 架构 | Vision Transformer |
| 输入 | 三维 FLAIR NIfTI |
| 病例聚合 | 25 张轴位有效切片的阳性概率均值 |
| 冻结阈值 | 0.548381 |
| 权重格式 | safetensors |
Quick Start
git lfs install
git clone <repository-url>
cd brain-tumor-flair-classifier
git lfs pull
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -e .
python examples\infer.py --input path\to\flair.nii.gz --output outputs\result.json
GPU 可用时默认使用 CUDA
python examples\infer.py --input path\to\flair.nii.gz --device cpu
Python API
from brain_tumor_flair_classifier import FlairClassifier
classifier = FlairClassifier.from_pretrained('本地模型目录')
result = classifier.predict_nifti('path/to/flair.nii.gz')
print(result['yes_probability'])
发布到 Hugging Face 后可直接使用模型 ID
classifier = FlairClassifier.from_pretrained('用户名/模型仓库名')
Output
{
"model_version": "v1",
"threshold": 0.548381,
"predicted_class": "yes",
"yes_probability": 0.721431,
"no_probability": 0.278569,
"evaluated_slices": 25
}
Evaluation
| 数据范围 | 分类结果 |
|---|---|
| 内部固定测试 | 29 / 30 正确 |
| UCSF-PDGM 外部开发阳性集 | 灵敏度 16 / 20 |
| HBN-SSI 外部开发健康集 | 特异度 12 / 12 |
| UPENN-GBM 最终盲测阳性集 | 灵敏度 9 / 10 |
| OpenNeuro ds003592 最终盲测健康集 | 特异度 10 / 10 |
| 合并最终盲测分类 | 19 / 20 正确 |
完整数据边界和评测说明见 docs/evaluation.md
Files
config.json模型配置model.safetensors微调权重preprocessor_config.json图像预处理配置examples/infer.pyNIfTI 推理示例NOTICE.md第三方组件与数据来源
License
代码采用 MIT License
- Downloads last month
- 19
Model tree for Songline/BrainTumor_FlairClassifier
Base model
google/vit-base-patch16-224-in21k