Evaluating and Mitigating Carbon Dioxide Equivalent Emissions in Stroke Management: A Modeling Study

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Environmental Science

Background:

• Artificial intelligence (AI) offers potential improvements in stroke imaging workflows.
• The environmental impact, specifically the carbon footprint of AI in healthcare, is not well understood.
• Estimating CO2 equivalent (CO2eq) emissions from AI in US stroke management is crucial for evaluating mitigation strategies.

Purpose of the Study:

• To quantify the carbon dioxide equivalent (CO2eq) emissions associated with AI use in US stroke imaging.
• To evaluate the effectiveness of carbon-aware mitigation strategies for AI in stroke care.

Main Methods:

• Utilized TriNetX neuroimaging data and Global Burden of Disease stroke estimates (2018-2019).
• Modeled two AI scenarios: Minimal AI (2D classification and segmentation models) and Ideal AI (multiple 2D/3D models).
• Calculated CO2eq emissions using state-level grid carbon intensity data.

Main Results:

• Ideal AI generated 16,375 metric tons CO2eq/year; Minimal AI generated 7,693 metric tons CO2eq/year.
• Computed tomography angiography and MRI were the primary contributors to emissions.
• Relocating processing to cleaner energy states reduced Ideal AI emissions by 54.75%.

Conclusions:

• AI-assisted stroke imaging possesses a quantifiable carbon footprint influenced by AI model complexity, imaging modality, and energy grid carbon intensity.
• Carbon-aware computation routing presents a viable immediate strategy to reduce environmental impact while enabling continued clinical AI adoption.