AI in Radiology: Enhancing Patient Care or Impeding Radiologist Performance?

AI in Radiology: Enhancing Patient Care or Impeding Radiologist Performance?

Artificial intelligence (AI) will change the way radiology is practiced. While AI has the potential to significantly improve patient care, there are also concerns that it may adversely alter radiologist behavior and productivity.

This blog post will explore three possible negative consequences of AI on radiologist behavior: alert fatigue, “cherry-picking,” and overreliance on AI leading to decision paralysis or complacency.

  1. Alert Fatigue: AI can help radiologists by providing automated preliminary findings or flagging potential abnormalities. However, an excessive number of alerts can lead to alert fatigue, where radiologists may become desensitized to these notifications. An analogous phenomenon has already occurred in inpatient hospital units, where it has been reported that 80% to 99% of alarms from monitoring devices are false or clinically insignificant. Constantly dealing with false positives or low-impact alerts will lead to radiologist fatigue and cause them to overlook critical findings or reduce their responsiveness to legitimate warnings. To mitigate this, AI algorithms need to be refined to minimize false positives and prioritize the most relevant alerts. Additionally, radiologists need to be trained on how to effectively manage alerts and how to identify false positives to maximize AI’s usefulness.

  2. Cherry-Picking Studies: Some radiologists may be tempted to avoid examinations on a work list that are flagged as abnormal by AI, especially if they perceive them as challenging or time-consuming. Cherry-picking radiology studies already exist as an issue, which AI may make worse. This phenomenon is potentiated by compensation models within radiology practices in which there is no financial incentive to read a potentially complex study versus a normal study. This behavior can lead to suboptimal patient care if important cases are overlooked and delayed. To address this, there should be a strong emphasis on encouraging radiologists to review all cases, regardless of AI results. This could be achieved through practice policy, automatic assignment to the next available radiologist by a smart work list, or by adjusting incentives for radiologists to prioritize and expedite the review of AI-flagged studies.

  3. Overreliance on AI and Complacency: There is a risk that some radiologists might come to rely too heavily on AI, trusting its decisions without critically evaluating the results. Introducing the AI diagnosis as the initial piece of information to the radiologist, even before they begin reviewing a study, runs the risk of increasing anchoring bias and overreliance on the technology. This was demonstrated in a study that showed that radiologist accuracy was significantly degraded when a computer-aided detection algorithm prompted radiologists with an incorrect diagnosis as opposed to the correct diagnosis. Furthermore, there could be a rise in instances of satisfaction of search errors, where radiologists might become content with the findings identified by the AI, leading them to overlook other crucial findings that the AI has not detected. These phenomena pose a particular risk to upcoming radiology trainees who will be exposed to AI while still in the process of learning radiology. To counter this, it’s essential to maintain continuous education and training for radiologists, emphasizing the importance of their expertise and the role of AI as an aid rather than a replacement.

To maximize the benefits of AI and minimize potential negative consequences, it is essential to integrate AI thoughtfully into the radiologist workflow and develop robust quality control measures. Regular feedback loops and assessments can help track AI performance, identify areas for improvement, and enhance radiologists’ ability to work effectively with AI.

Moreover, AI should be viewed as a tool to augment radiologists’ abilities rather than a direct substitute. By combining the strengths of AI with human expertise, radiologists can improve their accuracy, efficiency, and patient outcomes. As the technology continues to evolve, it will be vital to strike the right balance between AI and human skills to ensure the best possible patient care in radiology.

Picture of Varun Danda

Varun Danda

Varun Danda, MD, is an integrated interventional radiology resident at the Icahn School of Medicine at Mount Sinai, NY. He has a passion for medical technology, focusing on artificial intelligence, augmented reality, and medical devices in radiology.

Contact Us

CASE STUDY

ARA Health Specialists

Use the button below to download your free case study and learn how our approach to validation has improved the number of clinically significant findings in AI software.

CASE STUDY

Sutter Health

Use the button below to download your free case study and learn how our approach to validation has improved the number of clinically significant findings in AI software.