The rise of artificial intelligence (AI) has led to transformative changes in various fields, including radiology. While AI offers exciting possibilities, it also challenges ensuring fairness and equity in its implementation. As we navigate the complexities of AI, it is crucial to prioritize the responsible development and deployment of AI systems that ensure fairness, transparency, and inclusivity, thus promoting just and unbiased practices.
Equitable AI in radiology involves recognizing and rectifying biases in data collection, algorithm design, and deployment to avoid perpetuating existing disparities. Bias in radiology AI can arise from various sources, and all types of biases can translate into social biases, which can have adverse consequences for specific groups and exacerbate healthcare inequities. To ensure equitable AI, radiologists, and AI developers must work together to identify and mitigate bias through transparent and ethical practices, as the opaque nature of AI algorithms can hinder the understanding of the AI-generated results. AI models used in radiology should be designed to promote fairness and accountability with transparency and explainability in mind. Radiologists must have access to clear explanations for AI-driven findings.
Radiology AI also faces challenges in acquiring diverse and representative datasets. Transparency should also be maintained at the data level, and efforts should be made to compile comprehensive data that reflects the diverse patient population. Additionally, radiology departments should prioritize creating a culture of inclusivity and diversity to improve data collection practices.
Equitable AI implementation is particularly critical in resource-constrained settings, where healthcare disparities are more pronounced. In the context of global health and low-resource settings, the trustworthiness of AI for medical decision-making is hindered by several factors. Insufficient data diversity, coupled with nontransparent AI algorithms, creates barriers to achieving fairness and equity in these settings. Additionally, resource-poor health institutions encounter limitations regarding local equipment availability, personnel expertise, infrastructure development, data rights frameworks, and supportive public policies. So, they need a more holistic approach to the adoption of AI. Radiology organizations should support initiatives that provide education and infrastructure support for the responsible use of AI in such regions.
Radiology AI presents a transformative opportunity to improve patient care, and it is our commitment to fairness and inclusivity will shape a more just and equitable future for radiology. Promoting equitable AI in radiology requires collaboration among radiologists, AI developers, policymakers, and patients. Involving all stakeholders in the decision-making process ensures that AI systems address a variety of perspectives and avoid inadvertent biases. End-users of AI systems should be empowered and educated about potential sources of biases, enabling them to take proactive actions against bias and discrimination. This technology should be designed to be accessible and adaptable, considering the scarce resources in low- and middle-income countries, and should be monitored and evaluated continuously. Collaboration, forming diverse teams, and engaging and educating all stakeholders, including patients, in decision-making is the key to achieving equitable AI.
B. Bersu Ozcan, M.D. | Research Fellow, Breast Imaging Division | University of Texas Southwestern Medical Center