Although the use of artificial intelligence (AI) in medicine has a significant environmental impact, researchers report that it can still deliver net carbon benefits, particularly in radiology.
Kate Hanneman, MD, MPH, and colleagues at the University of Toronto report that data storage and computational efforts associated with AI are increasing their contribution to greenhouse gas emissions, particularly in radiology, which is a leader in healthcare AI. I wrote this in a review. Radiology.
However, AI can improve environmental sustainability in medical imaging by reducing MRI scan times, improving scanner scheduling efficiency, and optimizing decision support tools to reduce low-value image processing. , Hanneman and coauthors pointed out.
“AI applications can generate large amounts of greenhouse gases, primarily related to energy use, but on the flip side, they also have the potential to improve environmental sustainability,” Hannemann said. Told. today's med page. “This is the beginning of a conversation.”
AI contributes to greenhouse gas emissions in two ways, Hanneman said. First, the process of training, validating, and deploying AI models requires a large amount of computational power, which means more energy usage. Second, AI computing requires large amounts of data storage, increasing energy demands.
Computer rooms, server rooms and data warehouses “require a lot of energy to power and cool them,” Hanneman said.
While this is certainly true for radiology, there is a lack of data on specialty-specific impacts, such as energy estimation for computing and storing data specific to radiology tools, Hanneman said. .
“We make estimates and inferences based on non-radiological data, but to really understand what our actual energy use is and what the actual greenhouse gas emissions associated with it are… “Further research is greatly needed in this area,” Hanneman said. “There's a little bit of a gap there, so we need more data.”
At the same time, AI has the potential to create solutions for radiology that could have net benefits for the environment. For example, CT and MRI scanners consume significant amounts of energy even when idle, which AI tools can counter with automatic system shutdowns.
The researchers also wrote that because scan time is proportional to energy consumption, AI-based applications could help reduce MRI scan time and, in turn, energy consumption.
Both of these applications have the potential to reduce costs in radiology practices, while reducing energy use and greenhouse gas emissions, the researchers noted.
The authors suggest that radiology departments should optimize the energy use of AI models by adopting new approaches to technology implementation, such as the use of low-power processing equipment.
There's also the problem of “a proliferation of numerous AI models independently developed by different groups to address similar questions,” including a similar model for identifying pulmonary emboli in CT scans, the researchers said. is writing. “Efforts aimed at promoting cooperation and resource sharing will reduce overall greenhouse gas emissions associated with AI development, while also increasing the external validity of the resulting models,” they wrote. and is seeking cooperation from multiple institutions.
Additionally, data centers located in cooler climates or where renewable energy sources are available are more beneficial, so healthcare companies should “carefully evaluate their data storage choices to increase environmental sustainability.” “There is a need.
“Organizations that manage cloud storage can take immediate action by choosing sustainable partners,” Hanneman said in a statement.
While there are still gaps in knowledge about the impact of AI on radiology's environment, the goal is to open “a chapter in radiology in terms of thinking about both the positive and negative impacts of AI.” said Hanneman.
disclosure
The authors reported financial relationships with several companies and funding organizations.
Primary information
Radiology
References: Doo FX et al. “Environmental sustainability and AI in radiology: a double-edged sword” Radiology 2024; DOI:10.1148/radiol.232030.