Understanding the Environmental Impact of AI in Clinical Radiology
Data and energy-intensive artificial intelligence (AI) technologies are revolutionizing healthcare, particularly in clinical imaging, but often without regard to their environmental footprint. This article delves into a study aimed at assessing current practices concerning the environmental impact (ES) of AI-assisted pathways in radiology.
Methods and Analysis
To explore this, researchers conducted a comprehensive search of MEDLINE and Embase databases on November 5, 2024. They focused on quantitative clinical radiology studies published in English from 2015 onwards that employed machine learning techniques to support radiological diagnoses or interventions while discussing their ES implications. The primary outcome of interest was the quantitative reporting of environmental impacts, complemented by a secondary qualitative discussion within the text. The researchers synthesized these findings narratively and through an effect direction and magnitude analysis, excluding meta-analysis.
Results
Out of 4,449 records reviewed, 18 studies met the inclusion criteria. Notably, six of these studies (33.33%) provided quantitative ES results, and 15 (83.33%) offered a qualitative discussion on ES. Algorithms explicitly designed to be “lightweight” demonstrated a significant reduction in carbon emissions—ranging from 2.19 to 17.15 times (median 7.81) across 16 data points—and energy consumption, with a reduction factor of 1.60 to 751.62 times (median 3.22). These reductions were achieved without compromising clinical performance. However, no quantitative studies directly compared ES outcomes of AI-assisted pathways to standard care, and a notable 75% of studies discussing ES qualitatively did so with only a single sentence on sustainability.
Diploma and Recommendations
Despite growing concerns about AI’s climate impact, environmental outcomes remain an underexplored aspect in the evaluation of AI technologies in clinical radiology. The study reveals that there are indeed AI approaches with a lower carbon footprint. To align with sustainability objectives, it is crucial for funding bodies and governing institutions to foster the integration of environmental impact assessments into AI health research and development processes.
For those interested in further details, the study is registered under PROSPERO with the registration number CRD42024601818. To access the full study and additional insights, please visit the source link Here.
“`
