Why MRI and AI in Imaging is not Environmentally Sustainable and what can we do?

Brisbane Convention & Exhibition Centre 

Superconducting MRIs and the energy demands of Graphics Processing Units (GPUs) used in artificial intelligence (AI) development and deployment contribute significantly to greenhouse gas emissions and waste. MRI systems also depend on helium, a non-renewable resource derived from fossil fuels, while the compute requirements for supercomputers powering AI models—including the growing demands of training Foundation Models and large language models (LLMs) in imaging—further exacerbate environmental impacts. This talk will explore the potential role of AI and low-field or helium-free MRI in enhancing the sustainability of research and clinical imaging. We will discuss practical, easy-to-implement strategies, such as powering down MRI scanners when not in use, minimizing low-value imaging, adopting efficient low/zero-gadolinium protocols, and shortening MRI examination times. Additionally, we will consider how AI might transform lab environments by potentially eliminating the need for energy-intensive image viewing workstations if disease detection and diagnosis can be performed more accurately on raw and DICOM data by AI than by human interpretation.