Green AI for Accelerated medical imaging (GAIA)

Magnetic resonance imaging (MRI) is essential in medical diagnostics, providing detailed pictures of the body's internal structures. The integration of Artificial Intelligence (AI) has greatly enhanced MRI capabilities, improving diagnostic accuracy, speed, and efficiency. However, these advancements have also increased the carbon footprint of both MRI and AI applications. The GAIA project aims to reduce the carbon emissions associated with AI in medical imaging by developing innovative hardware and software technologies for data management. The health-care sector contributes 4-6% to global greenhouse gas (GHG) emissions. Medical imaging departments contribute, globally, up to 1% of GHG emissions and a single MRI scanner uses the same electricity as 26 four-person households. AI technology in medical imaging is making significant progress, for example by speeding up MRI scans, improving image quality, and enhancing efficiency. This helps reduce costs and energy use while allowing more patients to be scanned. However, AI benefits come with their own environmental burden. Developing AI for medical imaging is energy-intensive and increases GHG emissions due to big data storage and intensive model training. Although emissions linked to medical AI are currently modest (around 0.5 g of carbon for the analysis of a single patient's MRI scan), AI's environmental impact will grow as its implementation in practice inevitably expands. Given that over 95 million MRI scans are performed each year around the world, the large-scale use of AI in a single medical imaging application would contribute at least 52 tons of CO2 per year, equivalent to roughly 25 round-trip flights from Munich to New York. Multiple AI applications for each patient will raise this figure proportionally. Given the alarming conclusion from the Intergovernmental Panel on Climate Change (IPCC-2021) that global warming of 1.5-2 °C will be exceeded this century unless drastic reductions in CO2 and other GHG emissions are implemented, there is an urgent and unmet need for strategies to mitigate the environmental burden of AI and the informatics pipeline in medical imaging. GAIA will provide a sustainable solution for greener medical imaging by reducing the carbon footprint of AI-powered technologies. Its goal is to develop medical imaging technology that maintains high performance while minimising environmental and social costs. Instead of making AI systems more complex and energy-intensive, GAIA will shift the current AI paradigm towards developing smaller and simpler solutions. This will be done by developing both hardware and software solutions: affordable, low-energy devices running innovative lightweight AI models to handle MRI data, aiming for a nearly net-zero carbon footprint. Specific objectives are: Designing and optimizing lightweight deep learning-based models for MRI data storage, processing, analysis which achieve excellent performance keeping minimal model size and complexity. Designing and building affordable low-energy devices capable to train and deploy the new green AI models for sustainable and democratic medical imaging. Demonstrating the new GAIA technology on exemplar clinical and research-based MRI applications (e.g., brain lesions segmentation and characterization). With MRI imaging demand increasing by 2-8% annually, GAIA's efficiency will also give significant economic savings for the UK National Health Service and globally, e.g. reducing up to 75% the costs associated with handling medical imaging data (e.g., electricity consumption, cloud services, data storage/sharing). Although our initial focus is on MRI, we envision GAIA technology extending to other fields, reducing the carbon footprint of any resource- and energy-intensive data handling applications.