Software Engineer

GE Healthcare GE Healthcare · Healthcare · Beijing, Beijing, China · Digital Technology / IT

Software Engineer at GE Healthcare focused on building and training deep learning reconstruction networks (UNet/VarNet/physics-guided, score-based) using PyTorch/TF for CT applications. The role involves performance optimization, including GPU/CUDA acceleration and memory optimization for real-time processing, as well as engineering excellence in writing testable code, CI/CD, and containerization. It also includes integration with medical imaging standards and collaboration with physicists and clinicians for validation and productization.

What you'd actually do

  1. Implement CT application feature, Design, implement, and maintain application services and APIs
  2. Build and train reconstruction networks (UNet/VarNet/physics-guided, score-based) using PyTorch/TF
  3. Improve performance, reliability, observability, and developer tooling, GPU/CUDA acceleration, multithreading, memory optimization for real-time/near–real-time recon
  4. Write clean, testable code; unit/integration tests; CI/CD; containerization and deployment
  5. Work with DICOM/ISMRMRD; integrate with Gadgetron/BART/internal platforms

Skills

Required

  • PyTorch
  • TensorFlow
  • CT application feature development
  • API design and implementation
  • GPU/CUDA acceleration
  • real-time/near–real-time processing optimization
  • Clean, testable code
  • Unit/integration testing
  • CI/CD
  • Containerization
  • DICOM
  • ISMRMRD

Nice to have

  • UNet
  • VarNet
  • physics-guided models
  • score-based models
  • multithreading
  • memory optimization
  • Gadgetron
  • BART
  • web services
  • database tools
  • rules engines
  • Gradle
  • Maven
  • Git
  • SVN
  • Lean/Agile/XP
  • continuous integration (CI)

What the JD emphasized

  • Build and train reconstruction networks
  • GPU/CUDA acceleration
  • real-time/near–real-time recon

Other signals

  • Deep Learning
  • PyTorch/TF
  • GPU/CUDA acceleration
  • real-time/near–real-time recon