4 Ways AI-Based Medical Imaging Processing Can Be Improved
Medical imaging is one of the most complex and time-consuming tasks in medical diagnosis. AI-based imaging tools are designed to assist radiologists in reading medical images, but they are far from perfect. In this curated article, let’s explore four ways to make AI-based medical imaging better.
The use of artificial intelligence in medical imaging is not new. Infact computer-aided detection (CAD) tools has been in use for 25 years now to mark images of mammograms for possibility of breast cancer.
Though using CAD initially showed improved accuracy in detecting cancer, data from its wide adoption showed a much lower accuracy, that resulted in an increase in false-positive rates, overdiagnosis and unnecessary testing.
There are lessons to be learned by the result of premature adoption of CAD technology and now, an increasingly new AI-based imaging technologies’ rapid adoption into healthcare systems.
Keep The Human Touch
To combat the Automation Bias (the tendency of humans to defer to a more accurate computer algorithm if presented prior to a physician’s independent assessment) amongst physician using Artificial Intelligence guided medical imaging tools, we need to evaluate the human-computer-interaction (HCI) and interpreters for better understand how and when Artificial Intelligence outputs should be presented.
We will need to empower physicians to take the generated output from the Artificial Intelligence tool and evaluate that under the light of their expertise and experience to provide better diagnosis.
Value Outcome Rather Than Performance
The new Artificial Intelligence technologies in medical imaging need to demonstrate measurable and consistent outcomes in disease detection.
We need to prove that these new Artificial Intelligence tools are effective in improving the outcomes. We should study them in large, real-world screening settings with data collection and linkage to regional registries.
Even when more benefits than harms are identified and if those benefits are consistent across diverse populations and settings, they need to be confirmed to make sure that everyone is benefiting as much as possible.
Allow For Continuous Improvement
AI-based tools can improve their algorithms over time, which can be a big benefit. Sadly, current FDA review for AI tools in medical imaging is only provided for static, unchanging software tools.
One way to achieve this is to create prospectively collected imaging data sets that are representative of target populations and keep up with other temporal trends in medical imaging.
Define Standards to Mitigate Medical-Legal Risks
A great hope and promise of AI-based sophisticated algorithms in medical imaging is that it could eventually interpret images by themselves and give physicians more time to concentrate on more complex tasks. However, truly independent Artificial Intelligence is far from reality because physicians continue to be the responsible legal parties for accurate imaging interpretation.
One potential solution to reduce medical-legal risk is to better define standards and procedures as to who can interpret medical images. We need to define the how weightage needs to be assigned to an interpretation from the Artificial Intelligence tool from that of a medical radiologist.
In Conclusion
There is a widespread adoption of AI-directed tools in many areas of medicine beyond medical imaging, and they come with benefits and harms that we may not be aware of yet.
By adopting conscientious changes to our approaches in the new AI based tool’s physician interaction, outcomes, improvement, and the effort in reducing medical-legal risks, we can avoid repeating past mistakes.
Please follow my publication Digital Health Playbook for more stories.
Reference & Citation
Elmore JG, Lee CI. Artificial Intelligence in Medical Imaging — Learning From Past Mistakes in Mammography. JAMA Health Forum. 2022;3(2): e215207. doi:10.1001/jamahealthforum.2021.5207
