|
|
--- |
|
|
license: apache-2.0 |
|
|
library_name: transformers |
|
|
--- |
|
|
# MedVision-XRay |
|
|
|
|
|
<!-- markdownlint-disable first-line-h1 --> |
|
|
<!-- markdownlint-disable html --> |
|
|
<!-- markdownlint-disable no-duplicate-header --> |
|
|
|
|
|
<div align="center"> |
|
|
<img src="figures/architecture.png" width="60%" alt="MedVision-XRay" /> |
|
|
</div> |
|
|
<hr> |
|
|
|
|
|
<div align="center" style="line-height: 1;"> |
|
|
<a href="LICENSE" style="margin: 2px;"> |
|
|
<img alt="License" src="figures/badge.png" style="display: inline-block; vertical-align: middle;"/> |
|
|
</a> |
|
|
</div> |
|
|
|
|
|
## 1. Introduction |
|
|
|
|
|
MedVision-XRay is a state-of-the-art deep learning model for multi-label chest X-ray classification. The model has been trained on large-scale medical imaging datasets and can detect 14 different thoracic pathologies including pneumonia, atelectasis, cardiomegaly, and more. |
|
|
|
|
|
<p align="center"> |
|
|
<img width="80%" src="figures/performance.png"> |
|
|
</p> |
|
|
|
|
|
The model architecture is based on a modified ResNet backbone with attention mechanisms specifically designed for medical imaging applications. It achieves competitive performance across multiple medical imaging benchmarks. |
|
|
|
|
|
## 2. Evaluation Results |
|
|
|
|
|
### Comprehensive Benchmark Results |
|
|
|
|
|
<div align="center"> |
|
|
|
|
|
| | Benchmark | BaselineModel | ResNet-50 | DenseNet-121 | MedVision-XRay | |
|
|
|---|---|---|---|---|---| |
|
|
| **Detection Tasks** | Pneumonia Detection | 0.821 | 0.845 | 0.856 | 0.840 | |
|
|
| | Atelectasis Detection | 0.756 | 0.778 | 0.785 | 0.809 | |
|
|
| | Cardiomegaly Detection | 0.889 | 0.901 | 0.912 | 0.919 | |
|
|
| **Segmentation Tasks** | Lung Segmentation | 0.912 | 0.925 | 0.931 | 0.934 | |
|
|
| | Lesion Segmentation | 0.782 | 0.801 | 0.815 | 0.811 | |
|
|
| | Cardiac Segmentation | 0.845 | 0.862 | 0.871 | 0.861 | |
|
|
| **Classification Tasks** | Multi-label Classification | 0.723 | 0.751 | 0.768 | 0.775 | |
|
|
| | Severity Grading | 0.678 | 0.701 | 0.712 | 0.761 | |
|
|
| | Finding Localization | 0.645 | 0.678 | 0.692 | 0.688 | |
|
|
| **Robustness Tests** | Noise Resistance | 0.812 | 0.834 | 0.845 | 0.847 | |
|
|
| | Contrast Variation | 0.789 | 0.812 | 0.825 | 0.815 | |
|
|
| | Resolution Invariance | 0.756 | 0.778 | 0.789 | 0.800 | |
|
|
|
|
|
</div> |
|
|
|
|
|
### Overall Performance Summary |
|
|
MedVision-XRay demonstrates strong performance across all evaluated medical imaging benchmark categories, with particularly notable results in detection and segmentation tasks. |
|
|
|
|
|
## 3. Model Usage |
|
|
|
|
|
The model can be loaded using the Hugging Face Transformers library: |
|
|
|
|
|
```python |
|
|
from transformers import AutoModelForImageClassification, AutoFeatureExtractor |
|
|
|
|
|
model = AutoModelForImageClassification.from_pretrained("medical-ai/MedVision-XRay") |
|
|
feature_extractor = AutoFeatureExtractor.from_pretrained("medical-ai/MedVision-XRay") |
|
|
``` |
|
|
|
|
|
## 4. Training Details |
|
|
|
|
|
The model was trained with the following configuration: |
|
|
- Optimizer: AdamW with learning rate 1e-4 |
|
|
- Batch size: 32 |
|
|
- Image resolution: 224x224 |
|
|
- Augmentations: Random rotation, horizontal flip, color jitter |
|
|
- Training epochs: 50 |
|
|
|
|
|
## 5. License |
|
|
This model is licensed under the [Apache 2.0 License](LICENSE). For research and clinical use, please consult with your institution's IRB. |
|
|
|
|
|
## 6. Contact |
|
|
If you have any questions, please raise an issue on our GitHub repository or contact us at support@medvision-ai.org. |
|
|
|
