Breast Cancer Detection Tutorial

This tutorial demonstrates how to build a breast cancer detection pipeline using synthetic data augmentation with DiffuAug and distribution similarity evaluation with EvalDistSim.

Overview

Estimated time: 60 minutes
Difficulty: Advanced
Prerequisites: Python 3.9+, Familiarity with deep learning, Access to medical imaging data

What You’ll Learn

• Generate synthetic mammography data using DiffuAug
• Evaluate synthetic data quality with EvalDistSim
• Train a detection model with augmented dataset
• Compare performance against baseline

Dataset

This tutorial uses a sample mammography dataset. Due to data privacy concerns, we provide synthetic data for demonstration:

# Download sample dataset
wget https://example.com/sstdv-docs/sample-mammo.zip
unzip sample-mammo.zip -d ./data/

Step 1: Setup Environment

# Clone the repositories
git clone https://github.com/SSTDV-Project/DiffuAug.git
git clone https://github.com/SSTDV-Project/EvalDistSim.git

# Install dependencies
cd DiffuAug
pip install -e .
cd ../EvalDistSim
pip install -e .

Step 2: Generate Synthetic Data with DiffuAug

import diffaug
import numpy as np

# Load original dataset
original_data = np.load('./data/mammo_samples.npy')
labels = np.load('./data/mammo_labels.npy')

# Initialize DiffuAug
augmenter = diffaug.DiffuAug(
    model_type='ddpm',
    checkpoint='diffaug_pretrained.pth'
)

# Generate synthetic samples
n_synthetic = 500
synthetic_data = augmenter.generate(
    condition=labels,
    n_samples=n_synthetic,
    diffusion_steps=100
)

print(f"Generated {n_synthetic} synthetic mammograms")
print(f"Original shape: {original_data.shape}")
print(f"Synthetic shape: {synthetic_data.shape}")

Step 3: Evaluate Distribution Similarity

import evaldistsim

# Initialize evaluator
evaluator = evaldistsim.EvalDistSim()

# Compute distribution metrics
metrics = evaluator.compute_metrics(
    original=original_data,
    synthetic=synthetic_data,
    methods=['ks_test', 'wasserstein', 'fid']
)

print("Distribution Similarity Metrics:")
print(f"  Kolmogorov-Smirnov: {metrics['ks_test']:.4f}")
print(f"  Wasserstein Distance: {metrics['wasserstein']:.4f}")
print(f"  Fréchet Inception Distance: {metrics['fid']:.4f}")

Step 4: Train Detection Model

import torch
import torch.nn as nn
from torch.utils.data import DataLoader, TensorDataset

# Combine original and synthetic data
X_train = np.concatenate([original_data, synthetic_data], axis=0)
y_train = np.concatenate([labels, labels[:n_synthetic]], axis=0)

# Create dataloaders
train_dataset = TensorDataset(
    torch.FloatTensor(X_train),
    torch.LongTensor(y_train)
)
train_loader = DataLoader(train_dataset, batch_size=32, shuffle=True)

# Define simple detection model
class DetectionModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.features = nn.Sequential(
            nn.Conv2d(1, 32, 3, padding=1),
            nn.ReLU(),
            nn.MaxPool2d(2),
            nn.Conv2d(32, 64, 3, padding=1),
            nn.ReLU(),
            nn.AdaptiveAvgPool2d(1)
        )
        self.classifier = nn.Linear(64, 2)
    
    def forward(self, x):
        x = self.features(x)
        x = x.view(x.size(0), -1)
        return self.classifier(x)

# Training loop
model = DetectionModel()
optimizer = torch.optim.Adam(model.parameters(), lr=1e-3)
criterion = nn.CrossEntropyLoss()

for epoch in range(10):
    for batch_x, batch_y in train_loader:
        optimizer.zero_grad()
        outputs = model(batch_x)
        loss = criterion(outputs, batch_y)
        loss.backward()
        optimizer.step()
    print(f"Epoch {epoch+1}/10, Loss: {loss.item():.4f}")

Step 5: Evaluate Performance

# Evaluate on test set
model.eval()
with torch.no_grad():
    test_outputs = model(torch.FloatTensor(X_test))
    predictions = test_outputs.argmax(dim=1)
    accuracy = (predictions == y_test).float().mean()
    print(f"Test Accuracy: {accuracy:.4f}")

Expected Results

After completing this tutorial, you should have:

• 500 synthetic mammograms that pass distribution similarity tests
• A trained detection model with potential accuracy improvement from augmentation
• Quantitative metrics comparing original vs. synthetic data quality

Troubleshooting

Issue	Solution
Out of memory	Reduce batch size or use gradient checkpointing
Low FID score	Adjust diffusion steps or model architecture
Training instability	Reduce learning rate or increase warmup

Next Steps

• Brain Imaging Tutorial - Apply similar techniques to brain imaging
• EvalDistSim Reference - Deep dive into distribution evaluation
• DiffuAug GitHub - Latest updates and models

Citation

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

@article{sstdv2024breast,
  title={Synthetic Data Augmentation for Breast Cancer Detection},
  author={SSTDV-Project},
  year={2024}
}