Dataset Management

​

Overview

Sinkove datasets contain synthetically generated medical images created using advanced diffusion models. Each dataset is customizable and designed for research, AI model training, and educational purposes.

​

Dataset Types

​

Medical Imaging Modalities

Sinkove supports generation of various medical imaging types:

• Chest X-rays - Most common modality with multiple specialized models
• Other Modalities - Additional imaging types available based on model selection

​

Dataset Formats

All datasets are delivered as ZIP archives containing:

• Images - High-resolution synthetic medical images (typically 512x512 pixels)
• Metadata - JSON files with image annotations and generation parameters
• Model Cards - Documentation describing the AI models used for generation

​

Creating Datasets

​

Configuration Options

When creating a dataset, you can specify:

Model Selection

• Choose from available generative models optimized for different scenarios
• Each model may specialize in specific conditions or imaging characteristics

Prompts and Descriptions

• Use clinical terminology for accurate generation
• Examples: “Severe cardiomegaly”, “Bilateral pneumonia”, “Normal chest X-ray”
• Multiple prompts can be used to create diverse datasets

Dataset Size

• Currently limited to approximately 1000 images per dataset
• Larger datasets require more generation time

Quality Settings

• Configure generation parameters for your specific research needs
• Balance between quality and generation speed

​

Dataset Status Lifecycle

​

Status Phases

1. Pending - Dataset creation request is queued
2. Processing - AI models are actively generating images
3. Ready - Dataset is complete and available for download
4. Failed - Generation encountered an error
5. Expired - Dataset is no longer available (after retention period)

​

Monitoring Progress

Track your dataset status through:

• Web Dashboard - Real-time status updates in the platform
• Python SDK - Programmatic status checking
• API Endpoints - Direct REST API status queries

# Check dataset status
dataset = client.dataset("dataset_id")
print(f"Status: {dataset.status}")

# Wait for completion
while not dataset.ready():
    time.sleep(30)

​

Best Practices

​

Research Applications

Model Training

• Use diverse prompts to create balanced training sets
• Include both normal and pathological cases
• Consider augmenting real datasets with synthetic data

Validation Studies

• Generate datasets with known ground truth for algorithm testing
• Create edge cases that are rare in real clinical data
• Test model robustness with synthetic variations

Educational Use

• Create teaching datasets with clear diagnostic features
• Generate rare conditions for student training
• Provide consistent, high-quality examples

​

Technical Considerations

Data Management

• Download datasets promptly to avoid expiration
• Organize datasets with clear naming conventions
• Maintain metadata for reproducibility

Quality Assurance

• Review sample images before generating large datasets
• Validate synthetic images meet your research requirements
• Consider expert review for clinical accuracy

Ethics and Compliance

• Use synthetic data responsibly in research contexts
• Clearly label synthetic data in publications
• Follow institutional guidelines for AI-generated content

​

Storage and Retention

​

Download Requirements

• Immediate Download - Download datasets upon completion
• Retention Period - Datasets may be automatically deleted after a specified time
• Storage Limits - Account storage quotas may apply

​

Local Management

# Organize downloaded datasets
import os
from datetime import datetime

def organize_dataset(dataset_id, download_path="./datasets"):
    """Download and organize dataset with timestamp"""

    # Create organized directory structure
    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
    dataset_dir = os.path.join(download_path, f"{dataset_id}_{timestamp}")
    os.makedirs(dataset_dir, exist_ok=True)

    # Download dataset
    dataset = client.dataset(dataset_id)
    if dataset.ready():
        filepath = os.path.join(dataset_dir, "dataset.zip")
        dataset.save(filepath)
        print(f"Dataset saved to: {filepath}")

    return dataset_dir

​

Troubleshooting

​

Common Issues

Generation Failures

• Check prompt clarity and clinical accuracy
• Verify model compatibility with requested content
• Reduce dataset size if memory limitations occur

Download Problems

• Ensure stable internet connection for large files
• Verify sufficient local storage space
• Check API key permissions

Quality Issues

• Refine prompts with more specific medical terminology
• Try different models for better results
• Adjust generation parameters if available

​

Getting Support

For dataset-related issues:

• Review generation logs in your dashboard
• Contact support with specific dataset IDs
• Provide detailed descriptions of quality concerns