Digital pathology is at the forefront of medical innovation in the field of advanced diagnostic procedures, particularly through the use of artificial intelligence (AI). In the development of computationally and thus energy-intensive AI models, their sustainability is typically neglected in favor of performance. A team of scientists from the Institute of Pathology at Uniklinik RWTH Aachen has now developed a novel metric allowing a combined analysis of diagnostic performance with carbon footprint during model development. The researchers termed this metric “Environmentally Sustainable Performance score (ESPer)”.
The challenge: AI and its ecological footprint
AI models often require enormous computing resources, which is particularly true for AI models used in pathology. This leads to high energy requirements and associated CO₂ emissions, which vary depending on the energy source. So far, research has mainly focused on improving the diagnostic performance of AI models, while the ecological impact has hardly been considered. However, in the face of climate change, it is essential to implement resource-saving, sustainable strategies in the development and application of such technologies. In a previous publication, the same group has shown that the implementation of AI models in digital pathology can have considerable global warming potential.
Practical approaches to reducing the carbon footprint
To improve environmental sustainability, the researchers investigated strategies for reducing the CO₂ emissions of AI models. This included adjustments to the amount of data that needs to be analyzed by the AI models, e.g., by changing the image resolution and or reducing the analyzed image areas. These measures showed that it is possible to reduce energy consumption without significantly compromising diagnostic accuracy.
Why ecological sustainability matters in medicine
The cumulative impact of using resource-intensive AI models could significantly contribute to global climate change, especially if widely implemented in clinical practice. In addition, the large computing capacity required could be a limiting factor in widespread implementation. The ESPer score offers a new approach to help ensure that progress in medical AI does not come at the expense of the environment while maintaining or improving (diagnostic) performance. In this way, it also contributes to the United Nations Sustainable Development Goals (SDG 3: Good Health and Well-being and SDG 13: Climate Action).
A look into the future
The ESPer score is not only a tool for research, but also a potential standard for industry and regulatory authorities. It could serve as a guideline for sustainable AI development in the future. The researchers hope that their approach will spark further discussion and innovation at the intersection of technology and sustainability.
