AI Collaboration Cuts Energy Use in Medical Imaging

Researchers at an unspecified institution propose a new approach to federated learning that reduces energy consumption during the training of artificial intelligence models for converting MRI scans to CT scans. Federated learning allows multiple healthcare centers to collaboratively train AI without directly sharing patient data.

However, the process can be computationally expensive. This team developed a method that selectively “freezes” parts of the AI model during training, minimizing energy use without substantially impacting performance.

They tested the technique with five different federated architectures. The adaptive layer-freezing strategy reduced training time, total energy consumption, and carbon dioxide emissions by up to 23% compared to standard methods.

The accuracy of the MRI-to-CT conversion remained largely consistent, and even improved with some architectures. The work suggests a path toward more sustainable and equitable AI in healthcare, but further research could explore the long-term effects of layer freezing on model adaptability.