breast-cancer-swin-classifier

🩺 Breast Cancer Classification Using Swin Transformer (Hugging Face)

This repository contains a fine-tuned Swin Transformer Tiny model from Hugging Face trained on a colorized CT and X-ray breast cancer dataset for binary classification. The model classifies images into:

• Malignant (1)
• Benign / Normal (0)

The dataset includes colorized medical images enhanced with advanced colorization techniques to aid diagnostic AI models and medical professionals.

---

📊 Dataset Information

• Dataset Name: Medical Imaging (CT-Xray) Colorization New Dataset
• Author: Shuvo Kumar Basak (2025)
• Hosted on: Kaggle Dataset
• DOI: 10.34740/KAGGLE/DSV/11072909

Note: This dataset focuses on CT and X-ray images colorized with various methods to improve visual interpretation and enhance model training. The model does not include the dataset itself.

---

🧠 Model Architecture

• Model Type: Swin Transformer Tiny (pretrained on ImageNet)
• Framework: PyTorch + Hugging Face Transformers
• Base Model: microsoft/swin-tiny-patch4-window7-224
• Model Source: microsoft/swin-tiny-patch4-window7-224
• Input Size: 224x224 RGB
• Loss Function: CrossEntropyLoss
• Optimizer: Adam
• Fine-tuning: Final classification head modified for 2 classes

---

🏁 Training Pipeline Summary

1. Image Preprocessing:

   • Using Hugging Face AutoImageProcessor for normalization and resizing
   • Data augmentations via torchvision.transforms (horizontal/vertical flips, rotations, etc.)

2. Data Loading:

   • Dataset split into train, validation, and test sets
   • Loaded with torchvision.datasets.ImageFolder and wrapped in DataLoader

3. Model:

   • Loaded pretrained Swin Transformer Tiny from Hugging Face
   • Classification head adapted for binary classification (num_labels=2)

4. Training:

   • Optimizer: Adam
   • Loss: CrossEntropyLoss
   • Training on GPU

5. Saving:

   • Model weights saved as pytorch_model.bin
   • Hugging Face processor saved as needed

---

📈 Results

---

🔍 Intended Use

This model is intended for:

• Educational and research purposes in medical image analysis
• Building AI tools for breast cancer detection using colorized CT and X-ray images
• Demonstrating transfer learning with Hugging Face Swin Transformer models on medical datasets

⚠️ Not for clinical diagnosis or deployment without further validation and regulatory approval.

---

🚀 Inference Example (Python)

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

# Load model and processor
model_path = "Koushim/breast-cancer-swin-classifier"  # Your HF repo
processor = AutoImageProcessor.from_pretrained(model_path)
model = SwinForImageClassification.from_pretrained(model_path).eval()

# Load and preprocess image
image = Image.open("your_image.jpg").convert("RGB")
inputs = processor(images=image, return_tensors="pt")

# Predict
with torch.no_grad():
    outputs = model(**inputs)
    predicted_class = outputs.logits.argmax(-1).item()

labels = ["benign", "malignant"]
print("Predicted class:", labels[predicted_class])