Shivamnegi92/cancer-classification-vit

Cancer Image Classification using Vision Transformer (ViT)

Model Description

This model is a fine-tuned Vision Transformer (ViT) for binary classification of medical images to detect cancer. The model classifies images into two categories:

• Normal (Label 0): Healthy tissue
• Malignant (Label 1): Cancerous tissue

Model Details

• Base Model: google/vit-base-patch16-224-in21k
• Architecture: Vision Transformer (ViT-Base)
• Input Size: 224x224 pixels
• Number of Classes: 2 (Binary Classification)
• Framework: PyTorch + Transformers

Performance

The model achieves the following performance on the test dataset:

Metric	Value
Accuracy	83.38%
Precision	71.50%
Recall (Sensitivity)	72.25%
Specificity	88.02%
F1 Score	0.7187
MCC	0.6009

Clinical Interpretation

• ✅ Correctly classifies 83.4% of test images
• ⚠️ Catches 72.3% of actual cancer cases (27.75% false negative rate)
• ✅ 71.5% of cancer predictions are correct
• ⚠️ False Negative Rate: 27.75% - May miss some cancer cases
• ⚠️ False Positive Rate: 11.98% - May incorrectly flag healthy tissue

Usage

from transformers import AutoImageProcessor, AutoModelForImageClassification
from PIL import Image
import torch

# Load model and processor
model = AutoModelForImageClassification.from_pretrained("Shivamnegi92/cancer-classification-vit")
processor = AutoImageProcessor.from_pretrained("Shivamnegi92/cancer-classification-vit")

# Load and preprocess image
image = Image.open("path/to/your/image.jpg")
inputs = processor(images=image, return_tensors="pt")

# Make prediction
with torch.no_grad():
    outputs = model(**inputs)
    predictions = torch.nn.functional.softmax(outputs.logits, dim=-1)
    predicted_class = torch.argmax(predictions, dim=-1)

# Interpret results
class_names = ["Normal", "Malignant"]
confidence = predictions[0][predicted_class].item()
result = class_names[predicted_class.item()]

print(f"Prediction: {result}")
print(f"Confidence: {confidence:.4f}")

Training Details

Training Data

• Dataset: Private medical image dataset
• Training Samples: Custom split
• Validation Samples: Custom split
• Test Samples: 650 images (459 Normal + 191 Malignant)

Training Configuration

• Epochs: 5
• Batch Size: 16
• Learning Rate: 5e-5
• Optimizer: AdamW
• Loss Function: Cross Entropy
• Metric for Best Model: Accuracy

Data Processing

• Image Preprocessing: ViT image processor normalization
• Image Size: 224x224 pixels
• Data Augmentation: Random resized crop
• Label Mapping: Applied adjust_labels function (Normal=0, Malignant=1)

Limitations and Ethical Considerations

⚠️ Medical Disclaimer

This model is for research and educational purposes only. It should NOT be used for actual medical diagnosis.

Limitations

• High False Negative Rate: 27.75% of cancer cases may be missed
• Dataset Bias: Trained on specific dataset - may not generalize to all populations
• Image Quality Dependency: Performance varies with image quality and acquisition method
• Limited Scope: Only trained for specific type of medical imaging

Ethical Considerations

• Always consult qualified healthcare professionals for medical diagnosis
• Model outputs should be interpreted by medical experts
• Consider potential bias in training data
• Ensure proper validation in clinical settings before any medical application

Technical Specifications

Hardware Requirements

• Minimum RAM: 2GB
• Recommended RAM: 4GB
• GPU: Optional (CPU inference supported)
• Storage: ~400MB for model weights

Software Requirements

• Python: >=3.8
• PyTorch: >=2.0.0
• Transformers: >=4.46.0
• Pillow: >=9.0.0

Citation

If you use this model in your research, please cite:

@misc{cancer-classification-vit-2024,
  title={Cancer Image Classification using Vision Transformer},
  author={Shivam Negi},
  year={2024},
  howpublished={\url{https://huggingface.co/Shivamnegi92/cancer-classification-vit}}
}

Contact

For questions or issues regarding this model, please open an issue in the model repository.

---

Keywords: cancer detection, medical imaging, vision transformer, image classification, healthcare AI, diagnostic assistance