Advanced Medical Imaging AI Development
Budget: $30 – $250 USD
I need an AI engineer who can architect, train, and iterate on deep-learning models that perform both medical-imaging analysis and diagnostic support. The scope covers X-ray, MRI, and CT data, so you should be comfortable handling multimodal image pipelines and the differing pre-processing each modality demands.
You will start from a clean slate: selecting or designing network architectures in Python, building them with PyTorch or TensorFlow, and setting up a repeatable training environment that lets us experiment rapidly. Once a strong baseline is in place, I want to see steady, research-driven improvements—new loss functions, data-augmentation ideas, self-supervised techniques, or anything that reliably drives accuracy upward while keeping the models clinically robust.
Deployment matters as much as training. Please plan for containerised inference endpoints or lightweight on-prem solutions that radiology teams can plug straight into PACS/RIS workflows. Solid documentation, unit tests, and CI/CD hooks are expected so we can hand the code to hospital IT without surprises.
Deliverables:
• Clean, well-commented codebase with training scripts and reproducible environment files
• Trained weights for X-ray, MRI, and CT models plus a versioned model-registry structure
• Inference service (REST or gRPC) packaged for cloud or on-prem deployment
• Brief report summarising datasets, metrics (accuracy, sensitivity, specificity), and next research steps
If you thrive on continuous learning and can back your ideas with clear metrics, let’s build something genuinely useful for clinicians.
You will start from a clean slate: selecting or designing network architectures in Python, building them with PyTorch or TensorFlow, and setting up a repeatable training environment that lets us experiment rapidly. Once a strong baseline is in place, I want to see steady, research-driven improvements—new loss functions, data-augmentation ideas, self-supervised techniques, or anything that reliably drives accuracy upward while keeping the models clinically robust.
Deployment matters as much as training. Please plan for containerised inference endpoints or lightweight on-prem solutions that radiology teams can plug straight into PACS/RIS workflows. Solid documentation, unit tests, and CI/CD hooks are expected so we can hand the code to hospital IT without surprises.
Deliverables:
• Clean, well-commented codebase with training scripts and reproducible environment files
• Trained weights for X-ray, MRI, and CT models plus a versioned model-registry structure
• Inference service (REST or gRPC) packaged for cloud or on-prem deployment
• Brief report summarising datasets, metrics (accuracy, sensitivity, specificity), and next research steps
If you thrive on continuous learning and can back your ideas with clear metrics, let’s build something genuinely useful for clinicians.