Experts Renew Call for Patient-Centered Outcomes in AI Medical Imaging

November 7, 2024

Medical professionals are renewing their call for a shift in how artificial intelligence (AI) is utilized in medical imaging, emphasizing the need to focus on clinically meaningful outcomes rather than solely on the detection of radiographic abnormalities. This call to action builds upon insights from a seminal article published in The Lancet Digital Health by Dr. Ohad Oren, Professor Bernard J. Gersh, and Professor Deepak L. Bhatt.

Balancing Sensitivity with Clinical Relevance

AI technologies have demonstrated impressive capabilities in identifying subtle imaging changes, offering the potential for earlier and more precise diagnoses. However, experts caution that heightened sensitivity may lead to increased false positives and overdiagnosis. Detecting minor abnormalities without assessing their clinical significance can result in unnecessary anxiety for patients, additional testing, and unwarranted treatments.

“While AI can detect minute radiographic changes, not all of these findings require medical intervention,” the authors noted. “Emphasizing lesion detection without considering clinical impact may skew AI’s perceived performance and not necessarily improve patient outcomes.”

Challenges in Clinical Application

A primary concern is that AI algorithms might identify imaging patterns not easily recognized by clinicians, potentially complicating diagnosis and treatment. For instance, AI may detect early tissue changes in brain MRI scans suggestive of ischemic stroke, but the clinical significance of these subtle alterations remains uncertain. Without clear correlations to patient symptoms or long-term outcomes, such findings could hinder effective clinical decision-making.

In cancer screening, AI has shown higher sensitivity in detecting subtle lesions in mammograms. However, this increased detection does not always correspond to improved diagnostic accuracy over human radiologists and may lead to more false-positive results, subjecting patients to unnecessary procedures.

Recommendations for a Patient-Centered Approach

The experts advocate for strategies to align AI applications with patient-focused care:

• Prioritize Clinically Meaningful Endpoints: AI studies should focus on outcomes that directly impact patient health, such as symptom improvement, necessity for treatment, and survival rates, rather than merely detecting abnormalities.
• Refine AI Algorithms: Training AI systems to differentiate between benign and clinically significant lesions can reduce false positives and overdiagnosis, ensuring that detected abnormalities warrant medical attention.
• Enhance Validation Methods: Utilizing external validation and well-defined patient cohorts can improve the reliability of AI studies, facilitating their integration into clinical practice.

Potential Benefits When Applied Appropriately

When aligned with clinically meaningful endpoints, AI has the potential to significantly enhance patient care. In cardiology, for example, AI could help identify subtle cardiac changes predictive of adverse outcomes in conditions like aortic stenosis or myocarditis, enabling earlier interventions that improve survival and quality of life.

In oncology, AI’s ability to analyze complex imaging features may improve the detection and characterization of aggressive tumors while minimizing unnecessary interventions for indolent cases. For prostate cancer screening, AI could assist in distinguishing between tumors requiring immediate treatment and those suitable for active surveillance.

Continuing the Conversation

As AI technology advances, the medical community emphasizes the importance of integrating these tools thoughtfully to benefit patients truly. Ongoing research and dialogue are essential to ensure that AI applications enhance clinical outcomes without introducing new challenges.

About the Authors

• Dr. Ohad Oren is affiliated with the Division of Hematology and Oncology at Mayo Clinic, Rochester, Minnesota.
• Professor Bernard J. Gersh is part of the Department of Cardiovascular Medicine at Mayo Clinic College of Medicine, Rochester, Minnesota.
• Professor Deepak L. Bhatt is with the Brigham and Women’s Hospital Heart & Vascular Center and Harvard Medical School, Boston, Massachusetts.

References:

• Oren O, Gersh BJ, Bhatt DL. Artificial intelligence in medical imaging: switching from radiographic pathological data to clinically meaningful endpoints. Lancet Digital Health. 2020;2(9)–e488.